Old Change Logs

Revision: c17b548be6c25d08
Author: Evgeniy Fominov
Date: Aug 28, 2017 (18:00:17 UTC)

fix static runtime in cuda with clang and windows (#773)

---

Revision: 22ef457199658ec8
Author: Davis King
Date: Aug 27, 2017 (15:35:22 UTC)

Record last changeset and set PATCH version to 99

---

Revision: 9121e039950df935
Author: Davis King
Date: Aug 27, 2017 (15:34:24 UTC)

Created release v19.5

---

Revision: b2cc8f1da3f778d7
Author: Davis King
Date: Aug 27, 2017 (15:13:46 UTC)

Set this file to executable

---

Revision: 96e8daf12f37b8eb
Author: Guillaume Ramé
Date: Aug 27, 2017 (24:40:31 UTC)

improvements to cnn face detection python interface (#780) * improvements to cnn face detection interface * mmod rectangle object renaming. possibility to set batch size in multi image detection. Added check to make sure images are all the same size.

---

Revision: e7708c5dd4bc4d29
Author: Davis King
Date: Aug 27, 2017 (14:18:00 UTC)

Minor cleanup

---

Revision: c7d818690882a448
Author: Davis King
Date: Aug 27, 2017 (08:01:24 UTC)

Yet more comments

---

Revision: a245be12cb44c533
Author: Davis King
Date: Aug 27, 2017 (06:46:49 UTC)

cleanup

---

Revision: 00deb336c1cb6422
Author: Davis King
Date: Aug 27, 2017 (05:23:57 UTC)

Fixed compiler warning

---

Revision: df39ce20e7971294
Author: Davis King
Date: Aug 27, 2017 (05:23:49 UTC)

Fixed compiler warnings

---

Revision: 5e2a0b81bafb45f2
Author: Davis King
Date: Aug 27, 2017 (05:17:05 UTC)

Added youtube links

---

Revision: 3a9d5d4742ad6ad8
Author: Davis King
Date: Aug 27, 2017 (05:11:49 UTC)

More comments

---

Revision: bcf2eccb2563b3d9
Author: Davis King
Date: Aug 27, 2017 (04:29:36 UTC)

Added more comments

---

Revision: 3db195f39ddebe05
Author: Davis King
Date: Aug 27, 2017 (03:31:50 UTC)

Cleanup

---

Revision: 36f354fb6bba6236
Author: Davis King
Date: Aug 27, 2017 (03:31:38 UTC)

Gave upsample_image_dataset() an option to limit upsampling on really large images.

---

Revision: ff289ff45018c550
Author: Davis King
Date: Aug 26, 2017 (13:13:47 UTC)

Added two vehicle detection examples.

---

Revision: 9bc47990f9eea691
Author: Davis King
Date: Aug 26, 2017 (04:49:19 UTC)

Added a version of resize_image() that works inplace.

---

Revision: 01340662b0e82203
Author: Davis King
Date: Aug 26, 2017 (04:39:18 UTC)

Added overloads of max_pointwise() and min_pointwise() that take 3 arguments.

---

Revision: f723ce2ef3c754bc
Author: Davis King
Date: Aug 25, 2017 (18:29:40 UTC)

Made imglab --cluster ignore ignored boxes when doing all aspects of clustering.

---

Revision: 1f57c3504835dfba
Author: Davis King
Date: Aug 25, 2017 (18:01:02 UTC)

Added options to input_rgb_image_pyramid that let the user set create_tiled_pyramid()'s padding parameters. Also changed the default outer border padding from 0 to 11. This effects even previously trained models. So any model that doesn't explicitly set the outer patting to something else will have a padding of 11. This should be a more reasonable value for most networks.

---

Revision: 6a9b82d06a3afec8
Author: Davis King
Date: Aug 25, 2017 (16:35:47 UTC)

Gave create_tiled_pyramid() the ability to include padding around the outsides of the pyramid image.

---

Revision: 6418906f383c6b60
Author: Davis King
Date: Aug 25, 2017 (14:38:40 UTC)

Changed the functions that transform between input tensor coordinates and output tensor coordinates to use dpoint instead of point. This way, we can obtain sub-pixel coordinates if we need them.

---

Revision: f19f8301f63d9f18
Author: Davis King
Date: Aug 25, 2017 (14:21:35 UTC)

Made resize_image() and functions that use it like the pyramid objects produce better results when run on float and double images. There was needless rounding to integers happening in the bilinear interpolation. Now if you work with a float image the entire process will run without integer rounding.

---

Revision: 3eae6bb90b2bc710
Author: Davis King
Date: Aug 25, 2017 (14:20:17 UTC)

Clarified spec

---

Revision: 1cf38f0c65c0a297
Author: Davis King
Date: Aug 25, 2017 (08:59:55 UTC)

Clarified spec

---

Revision: e983c3baaffffb82
Author: Deniz Evrenci
Date: Aug 25, 2017 (27:40:22 UTC)

C++11 features (#778) * Make noncopyable constructor and destructor default C++11 provides the functionality. Defining empty functions cause all classes derived from noncopyable to be non-trivially constructible and non-trivially destructible. For example, matrix with compile-time layout by definition does not require an explicit destructor and should be trivially destructible ; however, deriving from noncopyable makes it non-trivially destrutible. This also affects vector<T, 2> and vector<T, 3>. * Delete array2d copy constructor and assignment operators

---

Revision: 8a4db312e35c76ba
Author: Deniz Evrenci
Date: Aug 25, 2017 (27:39:07 UTC)

Include drectangle.h to point_transforms.h (#777)

---

Revision: 7c97ae6e8fb6ccb2
Author: Davis King
Date: Aug 24, 2017 (15:42:22 UTC)

Fixed grammar

---

Revision: 37cc31f07f19f574
Author: Davis King
Date: Aug 24, 2017 (15:24:28 UTC)

Fixed spelling error in comment.

---

Revision: ed7d1cd9986c67b1
Author: Davis King
Date: Aug 24, 2017 (14:41:07 UTC)

Fully qualified boost::python::list to hopefully avoid compiler errors in some environments.

---

Revision: c99fbc13def29190
Author: Evgeniy Fominov
Date: Aug 24, 2017 (18:12:31 UTC)

Clang support for windows (#772)

---

Revision: 012387c880b9cb3e
Author: ipeterson
Date: Aug 23, 2017 (11:33:52 UTC)

Pull in external libpng dependencies properly (#770) PNG_LIBRARY set by libpng's FindPNG.cmake does not contain zlib dependancy. This causes the CHECK_FUNCTION_EXISTS(png_create_read_struct LIBPNG_IS_GOOD) to fail with liner errors, and for dlib to use it's internal copy of PNG. Updated to use libpng's PNG_LIBRARIES variable. This also sets both PNG and ZLib libraries in dlib_needed_libraries.

---

Revision: 68e4d543c0bc3316
Author: Davis King
Date: Aug 22, 2017 (18:02:19 UTC)

Fixed grammar

---

Revision: b5e05441d5e718fc
Author: Davis King
Date: Aug 21, 2017 (17:32:36 UTC)

Fixed incorrect size() for simd8i.

---

Revision: 44466b8bccd7eab7
Author: Davis King
Date: Aug 21, 2017 (17:27:38 UTC)

Made windows testing bat file use rmdir rather than rm since rm isn't always available on windows.

---

Revision: 28cdb9ba38ab7fe1
Author: Davis King
Date: Aug 21, 2017 (17:27:12 UTC)

Fixed linker errors when building pyhton on windows. This fixes a bug that was introduced in a recent PR. Also fixed compiler errors that occurred in visual studio.

---

Revision: 94fc5255cebe509f
Author: Davis King
Date: Aug 21, 2017 (15:35:42 UTC)

Fixed warning in visual studio.

---

Revision: 692a89abe25d650f
Author: Davis King
Date: Aug 20, 2017 (16:47:00 UTC)

Made the test loss in the verbose output messages from the dnn_trainer not jump in variance when the learning rate resets.

---

Revision: 4fd8c1322933d0f1
Author: Davis King
Date: Aug 20, 2017 (16:39:44 UTC)

Added serialization support for the running_stats_decayed object.

---

Revision: b2b9c59b46a25782
Author: Davis King
Date: Aug 20, 2017 (16:08:03 UTC)

Made the loss value management a little more conservative.

---

Revision: c3db550a43997eaf
Author: Davis King
Date: Aug 20, 2017 (15:42:12 UTC)

Made the input_tensor_to_output_tensor() and output_tensor_to_input_tensor() coordinate mappings work on networks that contain skip layers.

---

Revision: 4be725b8b96defea
Author: Davis King
Date: Aug 20, 2017 (15:30:11 UTC)

Changed the default get_test_iterations_without_progress_threshold() from 200 to 500. Now that we have a better history management of loss values in the trainer it's much more sensible to have a larger value here.

---

Revision: 4ce6f4a3e766e65a
Author: Davis King
Date: Aug 20, 2017 (15:28:08 UTC)

Made the dnn_trainer not forget all the previous loss values it knows about when it determines that there have been a lot of steps without progress and shrinks the learning rate. Instead, it removes only the oldest 100. The problem with the old way of removing all the loss values in the history was that if you set the steps without progress threshold to a really high number you would often observe that the last few learning rate values were obviously not making progress, however, since all the previous loss values were forgotten the trainer needed to fully populate it's loss history from scratch before it would figure this out. This new style makes the trainer not waste time running this excessive optimization of obviously useless mini-batches.

---

Revision: dee0369231464ef3
Author: Davis King
Date: Aug 20, 2017 (03:14:40 UTC)

The input_rgb_image_sized is supposed to be convertible to input_rgb_image, which it was in all ways except you couldn't deserialize directly like you would expect. This has now been fixed.

---

Revision: 935a4626e5a7650e
Author: Davis King
Date: Aug 19, 2017 (04:48:48 UTC)

Make DLIB_ASSERT statements not abort the python interpreter, but just trigger an exception.

---

Revision: 67eb90d2a5eab5f6
Author: Davis King
Date: Aug 19, 2017 (04:42:17 UTC)

Suppress compiler warning

---

Revision: e8b6ec9ee3de5dc2
Author: Adam Geitgey
Date: Aug 18, 2017 (06:30:33 UTC)

Add a python wrapper for using the mmod face detector (#753)

---

Revision: a5d7d71af997d1be
Author: Davis King
Date: Aug 15, 2017 (11:52:00 UTC)

Fixed spelling error

---

Revision: 22c842bb0ea24a87
Author: Davis King
Date: Aug 14, 2017 (10:04:16 UTC)

Updated solvers to correctly pull in cont_'s bias parameter multipliers.

---

Revision: 6fbcc1849eff3c47
Author: Davis King
Date: Aug 14, 2017 (09:50:47 UTC)

Added extract_ layer

---

Revision: 2735e30a69fde8ed
Author: Davis King
Date: Aug 14, 2017 (08:48:37 UTC)

Added more tests for copy_tensor()

---

Revision: 1786adb70faeead4
Author: Davis King
Date: Aug 14, 2017 (08:28:26 UTC)

Added an "add_to" option to tt:copy_tensor(). There was also a bug in the concat layer's backward() method. It was assigning the gradient to previous layers instead of adding the gradient, as required by the layer interface specification. This change also noticeably speeds up concat layers since only one CUDA kernel launch now happens per concat operation, rather than one kernel launch for each sample in a tensor.

---

Revision: 05ffcd2add434f40
Author: Davis King
Date: Aug 14, 2017 (05:52:53 UTC)

Made copy_tensor() use cudaMemcpyAsync() rather than cudaMemcpy().

---

Revision: 024940c678875ea6
Author: Davis King
Date: Aug 11, 2017 (13:47:19 UTC)

Added mult_prev layer.

---

Revision: 547865907036994d
Author: Davis King
Date: Aug 11, 2017 (12:39:00 UTC)

Added multiply_zero_padded()

---

Revision: f038970afe17f4ef
Author: Davis King
Date: Aug 10, 2017 (13:05:35 UTC)

Made swig always run when you rebuild to avoid stale swig outputs.

---

Revision: f9108f60012e7e6e
Author: Davis King
Date: Aug 10, 2017 (12:07:42 UTC)

Added DLIB_NO_ABORT_ON_2ND_FATAL_ERROR for dlib::fatal_error as a generic switch for use in plugin environments.

---

Revision: e9b7a2b23d9a9219
Author: Davis King
Date: Aug 10, 2017 (12:07:12 UTC)

Fixed missing java:: qualifiers.

---

Revision: 885c55c71009c06a
Author: Davis King
Date: Aug 10, 2017 (11:31:02 UTC)

Removed exit call on load library failure.

---

Revision: 0fc640f24e8baf64
Author: Davis King
Date: Aug 10, 2017 (11:16:21 UTC)

A bit of path cleanup

---

Revision: 1ff7211cc05b0422
Author: Davis King
Date: Aug 10, 2017 (10:38:41 UTC)

Added more options for controlling the install folder paths.

---

Revision: 7ded71367ca9d628
Author: Davis King
Date: Aug 18, 2017 (01:30:11 UTC)

Changed the random_cropper so that it samples background patches uniformly across scales regardless of the input image size. Previously, if you gave really large images or really small images it had a bias towards giving only large patches or small patches respectively.

---

Revision: 8b70bcb314ac92fc
Author: Davis King
Date: Aug 14, 2017 (17:36:18 UTC)

Fixed grammar

---

Revision: 039977decba9d23f
Author: Davis King
Date: Aug 09, 2017 (17:42:29 UTC)

merged

---

Revision: b2395d8b583f14d0
Author: Davis King
Date: Aug 09, 2017 (08:20:06 UTC)

merged

---

Revision: 0b26f900adfc416e
Author: Davis King
Date: Aug 09, 2017 (08:19:41 UTC)

Added an object that lets you hold a copyable reference to a java array. Also renamed the objects and generally improved documentation.

---

Revision: ac06c03e5ec8fc3a
Author: Davis King
Date: Aug 08, 2017 (11:10:17 UTC)

Added an overload of mat() that takes a row stride value.

---

Revision: 88ec0c153bcd9ff9
Author: Davis King
Date: Aug 08, 2017 (11:00:38 UTC)

Relaxes test to avoid false alarms.

---

Revision: 4e84498e368b7d95
Author: Davis King
Date: Aug 09, 2017 (17:42:14 UTC)

Fixed a bug in the warning message about NMS overlap where it would sometimes false alarm.

---

Revision: a3a15599f79256ed
Author: Davis King
Date: Aug 06, 2017 (07:34:07 UTC)

Added note about logging training parameters.

---

Revision: aea2c9f4a7daf579
Author: Davis King
Date: Aug 06, 2017 (04:54:10 UTC)

Changed how we print the network hash.

---

Revision: a2810bebf9f5a6bb
Author: Davis King
Date: Aug 05, 2017 (11:24:47 UTC)

Made dnn_trainer print the network size when logged to an iostream.

---

Revision: d10ac2240922c54a
Author: Davis King
Date: Aug 04, 2017 (18:46:46 UTC)

Improved how the relaxed mmod overlap settings are determined.

---

Revision: 65a32f50b0c67588
Author: Davis King
Date: Aug 03, 2017 (11:55:57 UTC)

Added get_synchronization_file() and get_test_one_step_calls() to dnn_trainer. Also added an operator<< for dnn_trainer that prints the parameters it's using. These changes also break backwards compatibility with the previous serialization format for dnn_trainer objects.

---

Revision: a72fd3b281342ddd
Author: Davis King
Date: Aug 03, 2017 (11:47:08 UTC)

Added operator<< for the DNN solvers.

---

Revision: a9e364c0ef45edec
Author: Davis King
Date: Aug 03, 2017 (11:40:51 UTC)

Added operator<< for random_cropper.

---

Revision: f333a561d3aa649b
Author: Davis King
Date: Aug 03, 2017 (11:28:26 UTC)

Filled out the options for loss_mmod's operator<<.

---

Revision: 6a1608b5d348414a
Author: Davis King
Date: Aug 03, 2017 (05:36:51 UTC)

Clarified spec

---

Revision: 6230ef7cabc2f565
Author: Davis King
Date: Aug 03, 2017 (05:14:36 UTC)

Fixed spelling error in comment.

---

Revision: c23b8fb76cd96163
Author: Davis King
Date: Jul 31, 2017 (15:49:46 UTC)

Added missing input validation to loss_mmod_. Specifically, the loss layer now checks if the user is giving truth boxes that can't be detected because the non-max suppression settings would prevent them from being output at the same time. If this happens then we print a warning message and set one of the offending boxes to "ignore".

---

Revision: 2d821a1b149b262b
Author: Davis King
Date: Jul 31, 2017 (15:43:49 UTC)

Clarified spec

---

Revision: 80e4c75351a5e37e
Author: Davis King
Date: Jul 30, 2017 (16:55:04 UTC)

Gave test_object_detection_function() an option to set how ignore box overlap is tested.

---

Revision: 49ad8b55b00ee788
Author: ᴚ∀W∀ᴎ
Date: Jul 29, 2017 (-05:42:42 UTC)

differential std::clamp from c++17 with dlib:: prefix to clamp() (#734)

---

Revision: b4c4c47cbeb4e6f9
Author: Leo
Date: Jul 28, 2017 (09:03:41 UTC)

Fix strange dash in Box-Muller (#732)

---

Revision: d9452d905bfb6ac6
Author: Davis King
Date: Jul 25, 2017 (11:42:13 UTC)

Updated cmake required version

---

Revision: 9a69628a43de0d97
Author: Davis King
Date: Jul 25, 2017 (11:41:30 UTC)

Added cmake scripts and some related tooling that makes it easy to call C++ code from java.

---

Revision: f1b45f1446062c59
Author: Davis King
Date: Jul 29, 2017 (01:51:59 UTC)

Minor adjustments to avoid false alarms in cca test.

---

Revision: 8fc176117752ac5f
Author: Davis King
Date: Jul 24, 2017 (06:01:30 UTC)

Fixed tabbing

---

Revision: 11755added24a539
Author: Davis King
Date: Jul 24, 2017 (06:01:23 UTC)

Fixed grammar

---

Revision: 69b6bb1130fc5fd3
Author: Davis King
Date: Jul 22, 2017 (07:41:48 UTC)

Improved errors from libpng when users link against a different version of libpng than they built against.

---

Revision: f7fbb3219aeaedc3
Author: Davis King
Date: Jul 22, 2017 (03:50:52 UTC)

Made tensor_conv hold references to the cuda_data_void_ptr work buffers in the member area of the class. This way, we avoid a potential error where the buffers are reallocated while cuDNN is still using them in the background.

---

Revision: d3ab7b9cda03f0de
Author: Evgeniy Fominov
Date: Jul 21, 2017 (22:37:46 UTC)

cuDNN convolution algorithms shared workspace (#695) * added shared workspace * rewrite shared workspace code * rename and device-based buffer allocation * fix cudnn_device_buffer constructors

---

Revision: 396b7ed0e6e08574
Author: Pavan kumar
Date: Jul 19, 2017 (23:68:28 UTC)

Just small typo (#716)

---

Revision: 98605f018fd15cc1
Author: Morosko
Date: Jul 18, 2017 (17:33:27 UTC)

Make http_client compile in C++11 and above (#714) * Make http_client compile in C++11 and above * Optimize file reading

---

Revision: bf5bfffb2b74e0b6
Author: Davis King
Date: Jul 18, 2017 (18:19:29 UTC)

Added --sort and also the ability to propagate boxes from one image to the next using dlib::correlation_tracker.

---

Revision: 4309baf01093704d
Author: Juha Reunanen
Date: Jul 16, 2017 (17:23:59 UTC)

Add per-pixel mean square loss (#690) * Add per-pixel mean square loss * Add documentation of loss_mean_squared_per_pixel_ * Add test case for per-pixel mean square loss: a simple autoencoder * Review fix: reorder params of function tensor_index, so that the order corresponds to the convention used in the rest of the dlib code base * Review fix: add breaks as intended, and change the rest of the test accordingly * Again a case where the tests already work locally for me, but not on AppVeyor/Travis - this commit is a blindfolded attempt to fix the problem (and it also fixes a compiler warning)

---

Revision: 6f82b11e5fc09b96
Author: Davis King
Date: Jul 15, 2017 (18:29:44 UTC)

Hopefully improved compiler error diagnostic when using gcc and a user misconfigures their build environment.

---

Revision: 89ab0bac484c0a62
Author: Joost van Amersfoort
Date: Jul 15, 2017 (03:12:00 UTC)

remove linking to libpython on linux/OSX (#703) * remove linking to libpython on linux * add OSX libpython free building * add automatic discovery of include python dir back in * make the libs non required for building on manylinux

---

Revision: 05f40d904621210b
Author: Davis King
Date: Jul 11, 2017 (18:17:07 UTC)

Trying another python build test.

---

Revision: 401c6e1e465f059d
Author: Davis King
Date: Jul 11, 2017 (18:02:26 UTC)

Trying again...

---

Revision: 72d0ff9885268f8c
Author: Davis King
Date: Jul 11, 2017 (16:46:29 UTC)

Add cmake to path for travis

---

Revision: 8cc5d90e0f1f70bb
Author: Davis King
Date: Jul 11, 2017 (15:52:18 UTC)

More python build tests

---

Revision: d0841d570136ae02
Author: Davis King
Date: Jul 11, 2017 (14:12:47 UTC)

Make travis-ci build the python api

---

Revision: fd0f8feaf3a1d09c
Author: MambaWong
Date: Jul 12, 2017 (10:11:06 UTC)

Update webcam_face_pose_ex.cpp (#702) Test on a given video like this cv::VideoCapture cap("Sample.avi") may be broken when the video frames are not enough before the main window is closed by the user.

---

Revision: a3a3d46411f8d79d
Author: Davis King
Date: Jul 09, 2017 (08:20:21 UTC)

Fixed typo in spec.

---

Revision: 24db262ee5fa81fc
Author: Davis King
Date: Jul 08, 2017 (12:53:29 UTC)

Added Juha Reunanen serialize test case, but one with dlib:: in front and one without to make sure both types of lookup work.

---

Revision: c2b2bc909b8f2acb
Author: Davis King
Date: Jul 08, 2017 (12:50:38 UTC)

Work around funny name lookup rules for serialize() call.

---

Revision: 49c1d15ec8edf4bc
Author: Davis King
Date: Jul 08, 2017 (11:56:40 UTC)

Fixed name lookup problem for calls to serialize() on network objects.

---

Revision: 2c9e5703a0c1c0bd
Author: Davis King
Date: Jul 07, 2017 (06:42:29 UTC)

Improved error message from the svmstruct object detector.

---

Revision: 8000a172af6d1a00
Author: Juha Reunanen
Date: Jul 07, 2017 (20:26:29 UTC)

Add new loss for weighted pixel inputs (#685) * Add new loss for weighted pixel inputs (may be useful e.g. to emphasize rare classes) * Deduplicate method loss_multiclass_log_per_pixel_(weighted_)::to_label * Add a simple test case for weighted inputs (also, fix a typo in test_tensor_resize_bilienar's name) * Add loss_multiclass_log_per_pixel_weighted_ to loss_abstract.h * Decrease the amount of weighting * There's no need to train for a very long time

---

Revision: 4233d5ba4a1aaad0
Author: Davis King
Date: Jul 06, 2017 (06:04:02 UTC)

Made ctrl+c detection in a mex file work more reliably in newer versions of matlab.

---

Revision: a7313c5603749aa2
Author: Davis King
Date: Jul 05, 2017 (08:12:43 UTC)

Renamed a few things to make the code and messages clearer.

---

Revision: 1d7f57b67c4421e3
Author: bot1131357
Date: Jul 06, 2017 (11:53:48 UTC)

Check for __ARM_NEON__ for libpng (#679) * Added check to see if __ARM_NEON__ is defined. Now we can use the following command: cmake --build --config Release .. * Rename to use_arm_neon.cmake to check_if_neon_available.cmake for clarity, minor tidying up of script, and simplifying try_compile() code for ARM NEON.

---

Revision: 97039314d589b354
Author: Davis King
Date: Jul 04, 2017 (09:19:41 UTC)

merged

---

Revision: 42edf220ec2b9da1
Author: Juha Reunanen
Date: Jul 03, 2017 (26:24:45 UTC)

Disable IntelliSense for all DNN unit tests (#678) * Problem: Visual Studio's vcpkgsrv.exe constantly uses a single CPU core, apparently never finishing whatever it's trying to do. Moreover, this issue prevents some operations like switching from Debug to Release (and vice versa) in the IDE. (Your mileage may vary.) Workaround: Keep manually killing the vcpkgsrv.exe process. Solution: Disable IntelliSense for some files. Which files? Unfortunately this seems to be a trial-and-error process. * Disable IntelliSense for the ResNet declarations * Disable IntelliSense for even more stuff * Disable IntelliSense for all DNN unit tests

---

Revision: e7500aaabbc0733b
Author: Davis King
Date: Jul 04, 2017 (09:14:05 UTC)

Added upsample_ layer that upsamples a tensor using bilinear interpolation.

---

Revision: a5d65a463f81c319
Author: Davis King
Date: Jul 04, 2017 (07:52:56 UTC)

Added tt::resize_bilinear() and tt::resize_bilinear_gradient().

---

Revision: 8b157ee76aa94558
Author: Davis King
Date: Jul 04, 2017 (04:52:12 UTC)

Clarified spec slightly.

---

Revision: 5104476d6c0f726c
Author: Davis King
Date: Jul 02, 2017 (04:43:45 UTC)

Added a comment

---

Revision: 9bb59097442a3284
Author: Davis King
Date: Jul 02, 2017 (04:16:16 UTC)

Removed default argument from process_batch since it upsets older versions of gcc and I can imagine scenarios where having it leads to user errors anyway.

---

Revision: 70e5e7dc8d677f68
Author: Davis King
Date: Jul 01, 2017 (12:26:58 UTC)

Moved label_to_ignore into loss_multiclass_log_per_pixel_ and also cleaned up a few minor things.

---

Revision: fad0fb2c55eee3d2
Author: Juha Reunanen
Date: Jul 01, 2017 (26:12:34 UTC)

Add new loss layer for semantic segmentation (pixel-wise classification) (#540) * #288 - add new layer loss_multiclass_log_matrixoutput for semantic-segmentation purposes * In semantic segmentation, add capability to ignore individual pixels when computing gradients * In semantic segmentation, 65535 classes ought to be enough for anybody * Divide matrix output loss by matrix dimensions too, in order to make losses related to differently sized matrices more comparable - note that this affects the required learning rate as well! * Review fix: avoid matrix copy * Review fix: rename to loss_multiclass_log_per_pixel * Review fix: just use uint16_t as the label type * Add more tests: check that network params and outputs are correct * Improve error message when output and truth matrix dimensions do not match * Add test case verifying that a single call of loss_multiclass_log_per_pixel equals multiple corresponding calls of loss_multiclass_log * Fix test failure by training longer * Remove the test case that fails on Travis for some reason, even though it works on AppVeyor and locally

---

Revision: 4dbaea3d09b01eb5
Author: Davis King
Date: Jul 01, 2017 (11:46:25 UTC)

Fixed broken copy constructor and assignment operator of con_, which became broken about an hour ago. Oops.

---

Revision: 222d67337366cc71
Author: Davis King
Date: Jul 01, 2017 (11:24:25 UTC)

suppress compiler warnings

---

Revision: c27b1f28cee43d94
Author: Juha Reunanen
Date: Jun 30, 2017 (15:47:21 UTC)

Problem: Visual Studio's vcpkgsrv.exe constantly uses a single CPU core, (#666) apparently never finishing whatever it's trying to do. Moreover, this issue prevents some operations like switching from Debug to Release (and vice versa) in the IDE. (Your mileage may vary.) Workaround: Keep manually killing the vcpkgsrv.exe process. Solution: Disable IntelliSense for some files. Which files? Unfortunately this seems to be a trial-and-error process.

---

Revision: 77d5852d5ca0e243
Author: Davis King
Date: Jul 01, 2017 (10:46:27 UTC)

Upgraded loss_mmod_ to support objects of varying aspect ratio. This changes the API for the mmod_options struct slightly.

---

Revision: 4d594cdbcfea8731
Author: Davis King
Date: Jul 01, 2017 (08:12:30 UTC)

Added process() and process_batch() to add_loss_layer. These routines let you easily pass arguments to any optional parameters of a loss layer's to_tensor() routine. For instance, it makes it more convenient to set loss_mmod_'s adjust_threshold parameter.

---

Revision: f46c094db0c8fbdb
Author: Davis King
Date: Jul 01, 2017 (07:54:20 UTC)

updated spec about con_

---

Revision: 1df4a8a67e958975
Author: Davis King
Date: Jul 01, 2017 (07:53:38 UTC)

Made it so you can set the number of output filters for con_ layers at runtime.

---

Revision: 307c1b509117864d
Author: Davis King
Date: Jul 01, 2017 (07:41:26 UTC)

Made dnn_trainer sync its state to two separate sync files that it alternates between. This should make syncing more robust to sudden hardware failure that happens right when saving to disk.

---

Revision: 5d7f0f9e1ac6f673
Author: Davis King
Date: Jul 01, 2017 (07:37:16 UTC)

Added select_oldest_file() and select_newest_file()

---

Revision: 93a30ab0bf7065d5
Author: Davis King
Date: Jul 01, 2017 (07:34:35 UTC)

updated docs

---

Revision: e44118d86e385fa9
Author: Davis King
Date: Jul 01, 2017 (07:20:45 UTC)

Added file::last_modified() for windows version of code.

---

Revision: b5b8b55a3d1c6781
Author: Davis King
Date: Jul 01, 2017 (06:38:15 UTC)

Added last_modified() method to dlib::file.

---

Revision: bc43b7b4a15064e7
Author: Davis King
Date: Jun 26, 2017 (17:06:59 UTC)

Cleaned up documentation for conv_. Also removed unnecessary tensor reallocation and copying inside conv_'s backward pass. Doing this required adding an add_to_output boolean option to the methods of tensor_conv.

---

Revision: 5f43b527f26068b6
Author: Davis King
Date: Jun 26, 2017 (17:01:47 UTC)

Fixed typo in comment.

---

Revision: d50bd56510b31ab6
Author: OranjeeGeneral
Date: Jun 22, 2017 (18:55:20 UTC)

refactored interface to reduce complexity so conv and convt layers forward passes have to call setup explicit now and there is only one ()-operator

---

Revision: e0b16a54ce28359e
Author: Joachim
Date: Jun 13, 2017 (14:33:42 UTC)

Merge branch 'master' of https://github.com/davisking/dlib

---

Revision: 224183b025f1b138
Author: Joachim
Date: Apr 10, 2017 (14:17:54 UTC)

Merge remote-tracking branch 'upstream/master'

---

Revision: 324a5638cdebfcb1
Author: Joachim
Date: Mar 16, 2017 (12:19:26 UTC)

fixed backward pass in cont layer to accumulate gradients this will pass the layer test now also removed compile warnings and changed some comments

---

Revision: 4e0b6915da3a4ddd
Author: Joachim
Date: Mar 13, 2017 (15:34:13 UTC)

add initial version of convolutional transpose layer

---

Revision: 88b121a1fedc1130
Author: Davis King
Date: Jun 28, 2017 (04:53:43 UTC)

Minor change to avoid errors from some versions of cmake.

---

Revision: e8cf0eeb0696abff
Author: Davis King
Date: Jun 21, 2017 (13:06:48 UTC)

Fixed mex class code printing

---

Revision: 9e977a1bed81dd2a
Author: Davis King
Date: Jun 21, 2017 (05:10:34 UTC)

Added operator= for simd8f so assignment from float compiles.

---

Revision: a871b6765e9f5461
Author: Grigoris
Date: Jun 07, 2017 (12:51:19 UTC)

Minor typo corrections in shape predictor trainer. (#633)

---

Revision: 03333233fd7f0e7f
Author: Plumtus
Date: Jun 07, 2017 (10:19:23 UTC)

Reinitialize averagers when saved sync file was reloaded. (#629)

---

Revision: 1b250933cc9382b3
Author: Davis King
Date: Jun 19, 2017 (16:54:45 UTC)

Made is so pressing e in imglab toggles between views of the image where the histogram is equalized or unmodified. This way, if you are looking at particularly dark or badly contrasted images you can toggle this mode and maybe get a better view of what you are labeling.

---

Revision: 722b93ed91244b64
Author: Davis King
Date: Jun 18, 2017 (04:11:54 UTC)

Updated code to work with new random_cropper interface.

---

Revision: 5faee3477597fea2
Author: Davis King
Date: Jun 17, 2017 (08:34:26 UTC)

Changed the random_cropper's interface so that instead of talking in terms of min and max object height, it's now min and max object size. This way, if you have objects that are short and wide (i.e. objects where the relevant dimension is width rather than height) you will get sensible behavior out of the random cropper.

---

Revision: 108467d3815cfcab
Author: Davis King
Date: Jun 06, 2017 (01:47:00 UTC)

Switched order of things in if statement so cmake hopefully won't give weird errors.

---

Revision: d2719abaf6546a6f
Author: Davis King
Date: Jun 04, 2017 (06:06:44 UTC)

Changed the converter so that, rather than producing one python file with everything in it, it now makes a python file as before but an additional binary file with all the weights in it. This way, if you are working with a network with a very large number of weights you won't end up with a crazy large python script.

---

Revision: ab209be410741632
Author: Davis King
Date: Jun 02, 2017 (18:50:01 UTC)

Added set_rect_area()

---

Revision: 5360858bb0c7f5e0
Author: Davis King
Date: May 31, 2017 (12:08:00 UTC)

Added a .fill() member to curand_generator that can create random 32bit integers.

---

Revision: 6021c01284a2bd82
Author: Evgeniy Fominov
Date: May 30, 2017 (15:40:11 UTC)

Possible CLang fix for Neon-based SIMD4i (#612)

---

Revision: 75122fdd0ac0cacb
Author: Davis King
Date: May 29, 2017 (16:06:37 UTC)

Relaxed the default non-max suppression parameters used by the mmod_options object so that users of the deep learning MMOD tool don't get spurious errors about impossibly labeled objects during training.

---

Revision: 232d654d576228ee
Author: Davis King
Date: May 28, 2017 (07:07:02 UTC)

Made the converter handle caffe's odd pooling layer output size calculations.

---

Revision: 033799853c74caf7
Author: Davis King
Date: May 27, 2017 (08:56:24 UTC)

Work around a bug in visual studio 2015.

---

Revision: 8d673c4363ac7494
Author: Davis King
Date: May 27, 2017 (07:59:32 UTC)

Made calling clean() on network objects also call clean on any layer detail objects that also provide a .clean() method.

---

Revision: 08f6d8abf9582e8b
Author: Davis King
Date: May 26, 2017 (13:23:58 UTC)

Added visit_layers_until_tag()

---

Revision: 96836695f84bbe15
Author: Juha Reunanen
Date: May 22, 2017 (18:58:27 UTC)

Noticed compiler warning C4800: 'double': forcing value to bool 'true' or 'false' (performance warning) (#538) - suggesting to change the test to what perhaps was the original intention

---

Revision: 525ffd8652c7e66b
Author: Davis King
Date: May 24, 2017 (03:24:12 UTC)

Made the converter add zero padding layers when needed by Eltwise to replicate the behavior of dlib's add_prev layers.

---

Revision: 7aead3cfc4860e72
Author: Davis King
Date: May 22, 2017 (15:11:00 UTC)

Made error message slightly better.

---

Revision: 0fa88f3431233d21
Author: Davis King
Date: May 22, 2017 (15:06:55 UTC)

Changed caffe converter to require the user to specify the input tensor size when the converter is run.

---

Revision: d975568a1a9e41cc
Author: emptyVoid
Date: May 20, 2017 (27:37:40 UTC)

Move shape_predictor_trainer to a separate header (#599) Moved shape_predictor_trainer to its own header in order to support the use of shape_predictor with DLIB_ISO_CPP_ONLY defined (i.e. without threading API wrappers).

---

Revision: 43a8742c0823d982
Author: Davis King
Date: May 19, 2017 (16:16:40 UTC)

Dlib and caffe actually do use max pooling layers with padding in the same way. So I just removed the error check that was preventing the conversion from proceeding in that case. I also added more useful output messages about setting input tensor dimensions.

---

Revision: cbeec8729b42008c
Author: Davis King
Date: May 17, 2017 (12:43:32 UTC)

Fixed uninitialized variable in test that caused unreliable test results.

---

Revision: d3fd1a0bc906b749
Author: Davis King
Date: May 16, 2017 (17:14:09 UTC)

Cleanup and also fixed a cmake error when building dlib outside a project.

---

Revision: 8a628dc5a9c53d8a
Author: tschw
Date: May 16, 2017 (22:47:53 UTC)

Fix installed CMake targets when DLIB_ISO_CPP_ONLY (#595) (#597)

---

Revision: 82781a9eb9329af9
Author: Davis King
Date: May 16, 2017 (16:50:34 UTC)

yet more print statements

---

Revision: 0ec5f02648fba710
Author: Davis King
Date: May 16, 2017 (05:55:34 UTC)

A test to see more info about what's going weird on appveyor.

---

Revision: fd29101367974642
Author: Davis King
Date: May 15, 2017 (03:12:18 UTC)

More logging messages

---

Revision: 8e0cabe86d7e501d
Author: Davis King
Date: May 14, 2017 (16:07:16 UTC)

A minor change to hopefully reduce inane warnings from visual studio.

---

Revision: a48bc42b6ba55508
Author: Davis King
Date: May 14, 2017 (15:55:01 UTC)

removed cruft

---

Revision: 4a38ad612a5dae29
Author: elelel
Date: May 14, 2017 (23:52:34 UTC)

Migrate from dlib::scoped_ptr to std::unique_ptr (#593) * Convert unique_ptr * Fix passing unique_ptr as copy by value * Remove scoped_ptr implementation * Fix missed files * Move bool cast into tester macro * Reexport scoped_ptr alias from sockets

---

Revision: ef0782f1c88c775c
Author: elelel
Date: May 13, 2017 (22:05:52 UTC)

Change type traits from inherited to explicit (#591)

---

Revision: dc1e020634471405
Author: Davis King
Date: May 13, 2017 (10:23:24 UTC)

Added more outputs to unit test error message.

---

Revision: da3e6e946e500700
Author: elelel
Date: May 13, 2017 (18:05:23 UTC)

Replace shared_ptr_thread_safe with std::shared_ptr (#590) * Replace shared_ptr/weak_ptr with stdlib counterparts * Fix ptr usage through tests compilation * Bring back dlib smart ptrs as legacy * Include scoped_ptr directly * Add explanation about smart_ptr deprecation * Replace shared_ptr_thread_safe * Fix missed old include * Removed missed old include

---

Revision: f1ba17502aeafe99
Author: elelel
Date: May 12, 2017 (20:40:59 UTC)

Replace shared_ptr/weak_ptr with STL's versions (#588) * Replace shared_ptr/weak_ptr with stdlib counterparts * Fix ptr usage through tests compilation * Bring back dlib smart ptrs as legacy * Include scoped_ptr directly * Add explanation about smart_ptr deprecation

---

Revision: 073194d2bf33b2a5
Author: Davis King
Date: May 07, 2017 (15:40:42 UTC)

Fixed spelling error

---

Revision: 7c5ef62bc9326d82
Author: Davis King
Date: May 07, 2017 (15:12:12 UTC)

Added a note about exporting to XML.

---

Revision: dd4d30f3521efe26
Author: Davis King
Date: May 07, 2017 (15:11:52 UTC)

Changed the location of the set_synchronization_file() call so that rerunning the example restarts correctly.

---

Revision: 755c0ac22b15ee88
Author: Davis King
Date: May 07, 2017 (15:08:41 UTC)

Renamed file

---

Revision: 0a24828d2080d8a8
Author: Davis King
Date: May 07, 2017 (15:07:18 UTC)

Improved example python script

---

Revision: a950eb8d61bb2530
Author: Davis King
Date: May 07, 2017 (13:16:08 UTC)

Added more comments

---

Revision: d66e5da9d57d2aa0
Author: Davis King
Date: May 07, 2017 (12:57:34 UTC)

Fixed affine_fc and clarified error message.

---

Revision: d73d2d82bb409fb3
Author: Davis King
Date: May 07, 2017 (12:23:35 UTC)

Added missing sig and prelu layer support.

---

Revision: e1dffb71454fee96
Author: Davis King
Date: May 07, 2017 (11:49:19 UTC)

Cleaned up how the output is output.

---

Revision: e414011a9bbd73d1
Author: Davis King
Date: May 07, 2017 (11:26:18 UTC)

Added support for all the other dlib layers that make sense.

---

Revision: b422f960daf58c78
Author: Davis King
Date: May 06, 2017 (11:02:38 UTC)

Added a simple tool to convert dlib model files, which are saved as XML files, to caffe models. Not all layers are supported yet, but it works on the net from the first dlib dnn example program. Will add the rest of the layers soon.

---

Revision: 93ddd177d4241cda
Author: Davis King
Date: May 07, 2017 (13:32:15 UTC)

Added an overload of net_to_xml() that takes a string filename.

---

Revision: 0deaaacf074903fe
Author: Davis King
Date: May 07, 2017 (13:19:26 UTC)

Changed the XML format output by net_to_xml(). Specifically, the XML tag for affine layers was changed so to use the same conventions as other layers that support convolutional vs fully connected modes.

---

Revision: e326d0c0a08262ec
Author: Davis King
Date: May 07, 2017 (13:17:56 UTC)

Made net_to_xml() output floating point numbers with more digits.

---

Revision: b1b5e6de383df644
Author: Davis King
Date: May 06, 2017 (08:46:19 UTC)

Made the attribute_list of the xml parser a little more friendly by allowing you to ask for attributes that don't exist and get a defined behavior (an exception being thrown) rather than it being a contract violation.

---

Revision: 140a70ec4fc94561
Author: Davis King
Date: May 05, 2017 (01:42:13 UTC)

Added more comments

---

Revision: 25cf9e1177db1f08
Author: Jim61C
Date: May 03, 2017 (01:40:04 UTC)

remove enum in function signature to avoid cblas declaration conflicts (#571)

---

Revision: a7859342b893164d
Author: Evgeniy Fominov
Date: May 01, 2017 (21:31:36 UTC)

Multiarch installation fix (#567) * fix multiarch install * fix include dir * cleaned up stuff and removed things we don't need anymore

---

Revision: 16134dee16e1933e
Author: cclough
Date: Apr 30, 2017 (10:48:51 UTC)

Update CMakeLists.txt (#566)

---

Revision: cecc99498a2323bb
Author: Evgeniy Fominov
Date: Apr 29, 2017 (25:39:35 UTC)

ARM NEON SIMD support (#564) * added neon functions * TK1 tests passed * code cleanup * Re-tested on TK1 * improve simd4i * fixed simd4f_bool neon * fix simd8i sse operator <= * restored rsqrt/VSX * fixed simd4f/neon/reciprocal_sqrt

---

Revision: 7c55b1ef14be9a97
Author: Davis King
Date: Apr 21, 2017 (06:45:02 UTC)

Added tt::log(), tt::exp(), and tt::log10()

---

Revision: f96f0a6d552bf009
Author: Davis King
Date: Apr 21, 2017 (06:15:27 UTC)

Fixed tabbing

---

Revision: 08eca19334f45ac4
Author: David Seifert
Date: Apr 16, 2017 (20:14:21 UTC)

CMake improvements (#539) * Use GNUInstallDirs to allow changing target directories * Multi-lib distributions need to change the libdir which is only portably possible with `GNUInstallDirs`. The current `LIB_INSTALL_DIR` solution is not portable and not supported upstream by Kitware. * Multi-arch distributions such as Debian and Exherbo need to be able to change the bindir. * Do not install LICENSE.txt * Most distributions do not install license files, as they have all licenses stored in a central repository, in order not to duplicate the same file over and over. * License files do not need to be installed for a piece of software to be licensed under that license.

---

Revision: 4157beda1498593a
Author: Roy Macdonald
Date: Apr 14, 2017 (18:41:22 UTC)

Fixes issue with object named nil that collide with apple's definition. (#537) * Fixes issue with object named nil that collide with apple's definition. * fix nil named objects: changed name to no_parent

---

Revision: 17edc910ea074b59
Author: ernestotapiar
Date: Apr 14, 2017 (17:41:25 UTC)

Corrected typo that generated an erroneus Hessian of Brown's function. (#534)

---

Revision: 8a6bf5db9bc39fd4
Author: Joost van Amersfoort
Date: Apr 07, 2017 (17:02:37 UTC)

Add support for Python3.6 lib search with Cmake (#527) * Add support for Python3.6 lib search with Cmake A recent commit added support for python3.5 lib search with Cmake. This PR extends that to python 3.6. Relevant old commit: https://github.com/davisking/dlib/commit/17a2c75ddffb13e915686803c5e3bf4cc1389a96 * Turn python versions into a list

---

Revision: 81942f1a2e279806
Author: Davis King
Date: Apr 19, 2017 (18:19:17 UTC)

Added python requirements.txt file for scikit-image

---

Revision: 8f7170dcdb5139c5
Author: Davis King
Date: Apr 06, 2017 (16:28:08 UTC)

Tweak assert enable/disable behavior so the asserts still come on or off as a function of the release/debug drop down menu when using visual studio's IDE.

---

Revision: 5739b256a6eb6138
Author: Martijn Courteaux
Date: Apr 07, 2017 (03:41:20 UTC)

Fix asserts, super clean! (#526) * Wether asserts are enabled or not, is now only suggested by dlib, and not longer imposed. The user can override using SET(CACHE FORCE) or command line -DDLIB_ENABLE_ASSERT=ON/OFF anytime. * Fixed missing line in cmake for asserts.

---

Revision: 77669010b5182f50
Author: Davis King
Date: Apr 06, 2017 (13:07:07 UTC)

Stopped using CMAKE_CXX_FLAGS to set dlib's preprocessor switches and instead now use target_compile_options(). This should avoid weird configuration errors that happen in rare instances on some platforms.

---

Revision: db665c35d1076ed4
Author: Davis King
Date: Apr 05, 2017 (16:02:26 UTC)

Fixed cmake so openmp is handled properly in visual studio. The change yesterday messed it up and effectively disabled cuda in visual studio. This fixes that.

---

Revision: dc709769b48049ba
Author: Davis King
Date: Apr 05, 2017 (07:50:20 UTC)

Minor CMake cleanup

---

Revision: f5255d583370c41d
Author: Davis King
Date: Apr 05, 2017 (07:19:00 UTC)

Don't try to use any CUDA stuff when making matlab mex files.

---

Revision: f4ed7e576481801f
Author: Davis King
Date: Apr 05, 2017 (07:08:32 UTC)

Fixed bad results when using CUDA and the Intel MKL at the same time. This was because of a bug introduced yesterday in the CMake scripts which would accidentally pull in two versions of the OpenMP libraries, the GNU one and the Intel one, but you can't mix them together at the same time.

---

Revision: 3f189157af659cb9
Author: goodloop
Date: Apr 05, 2017 (30:58:15 UTC)

fix unneeded reference for cblas_strsm and cblas_dtrsm when DLIB_USE_BLAS is not defined (#519)

---

Revision: 6ead4fdd4cdbcae9
Author: Davis King
Date: Apr 05, 2017 (02:15:39 UTC)

Fixed visual studio errors about openmp compiler flags not being right.

---

Revision: dbc83cda3b5401e3
Author: Davis King
Date: Apr 05, 2017 (01:47:06 UTC)

Added missing quotes so that cmake works on windows.

---

Revision: 7836cc0c0b216607
Author: Davis King
Date: Apr 04, 2017 (17:37:38 UTC)

Fixed code not compiling when cuda is disabled.

---

Revision: 1d99d801574d5434
Author: sigbjorn
Date: Apr 04, 2017 (22:28:19 UTC)

Provide cleanup for timer_global_clock singleton, issue #505 (#512) * temporary workaround for python hanging on dlib::shared_ptr_thread_safe<global_clock> when extension-module using dlib is unloaded on windows platform * added some more doc and tests for the new delete_global_clock function. Minor adjustment to singleton pattern to allow easy access to singleton shared ptr * alternate _WIN32 specific solution, with no extra functions nor user-actions needed * removed floating/obsolete delete_global_clock, tab to space * leaving linux code unchanged in static desctructor * cleanup not modified files * align formatting

---

Revision: 12e38fe71aeed300
Author: Davis King
Date: Apr 04, 2017 (17:18:30 UTC)

Added a tt::inv() for computing matrix inversions on the GPU.

---

Revision: 154c991285d2c6ff
Author: Davis King
Date: Apr 04, 2017 (11:56:58 UTC)

removed cruft

---

Revision: d684b7468c8f2bfa
Author: Davis King
Date: Apr 04, 2017 (11:21:19 UTC)

Moved the cuda_data class into its own file, renamed it to cuda_data_ptr, and also pushed all the cuda #includes into a .cpp file.

---

Revision: d0562135ef97eace
Author: Davis King
Date: Apr 04, 2017 (10:30:06 UTC)

Added cusolver_error class.

---

Revision: c7e95f3b5f375a5b
Author: Davis King
Date: Apr 04, 2017 (10:29:41 UTC)

Added cuda_data, a templated container for GPU memory.

---

Revision: c191fe1994525a0a
Author: Davis King
Date: Apr 03, 2017 (02:10:19 UTC)

Added support for cuDNN 6

---

Revision: 91ce4560b9328fb4
Author: Davis King
Date: Apr 02, 2017 (12:31:55 UTC)

fixed spelling error in comment

---

Revision: e4468f6d706476df
Author: Davis King
Date: Apr 02, 2017 (10:03:51 UTC)

Added a version of tt::affine_transform() that operates on a sub-rectangle rather than the entire tensor.

---

Revision: 4d29d54c03ae914c
Author: Davis King
Date: Mar 30, 2017 (06:43:49 UTC)

Made the title bar show the current image's number.

---

Revision: dcd6d464fa69c862
Author: marcovzla
Date: Mar 24, 2017 (07:20:33 UTC)

remove miniconda (#500)

---

Revision: 8f7255877534648a
Author: Davis King
Date: Mar 30, 2017 (03:32:42 UTC)

Fixed the visit_layers_backwards() and visit_layers_backwards_range() routines so they visit layers in the correct order.

---

Revision: 8b28b30a98d91f99
Author: Davis King
Date: Mar 27, 2017 (17:00:37 UTC)

Changed from ctrl to alt because otherwise it's easy to accidentally hit ctrl up and jump when you don't meant to while pruning an image list.

---

Revision: 386305577dc83e18
Author: Davis King
Date: Mar 27, 2017 (16:13:39 UTC)

Changed to ctrl+d since it's easier to press.

---

Revision: 6154ad414637b54a
Author: Davis King
Date: Mar 27, 2017 (16:11:04 UTC)

Made it so you can remove images by pressing ctrl+r.

---

Revision: 22568dfb174282f5
Author: Davis King
Date: Mar 24, 2017 (05:15:43 UTC)

Made it more obvious that this file is a cmake tutorial. Also added a few more comments about how to use cmake.

---

Revision: e0aad42281e637fa
Author: Davis King
Date: Mar 22, 2017 (11:59:53 UTC)

merged

---

Revision: 465eff072b9d31fa
Author: Davis King
Date: Mar 22, 2017 (11:59:06 UTC)

Added another code path inside solve_qp_box_constrained_blockdiag() that is much faster when the off-diagonal vectors are all simple multiples of the ones_matrix().

---

Revision: a1458a5153637fe1
Author: Davis King
Date: Mar 22, 2017 (08:49:12 UTC)

Added missing requires clause.

---

Revision: 76cb31c027c18564
Author: Davis King
Date: Mar 22, 2017 (07:00:27 UTC)

Made the mex wrapper deal with cell arrays that have null elements.

---

Revision: bbcdc96dc4fb68e2
Author: Davis King
Date: Mar 21, 2017 (17:05:21 UTC)

Added age and gender fields to image metadata.

---

Revision: 26b9c3127cecd89d
Author: Davis King
Date: Mar 21, 2017 (05:50:01 UTC)

Removed unneeded code.

---

Revision: d4bf051a4dd87384
Author: ebroglio
Date: Mar 18, 2017 (21:54:46 UTC)

dets cleared twice in evaluate_detectors of scan_fhog_pyramid.h (#487) already cleared at the beginning

---

Revision: b11d0851a3019b42
Author: ebroglio
Date: Mar 18, 2017 (21:17:52 UTC)

remove unused variable in edge_list_graphs.h (#486)

---

Revision: d6e70cc1c54f7f57
Author: Davis King
Date: Mar 20, 2017 (17:10:36 UTC)

Fixed error in pyramid_down<2> that caused the output image to be a little funny looking in some cases.

---

Revision: 29e864121dcf27dc
Author: Davis King
Date: Mar 20, 2017 (12:41:22 UTC)

Added solve_qp_box_constrained_blockdiag()

---

Revision: d290c16ba7d36831
Author: Davis King
Date: Mar 18, 2017 (09:00:16 UTC)

Fixed double delete

---

Revision: 7100036ec519f6aa
Author: Davis King
Date: Mar 16, 2017 (07:30:02 UTC)

Removed use of const_temp_matrix here because it prevents binding to BLAS.

---

Revision: 05aded11b02710a3
Author: Davis King
Date: Mar 09, 2017 (15:53:35 UTC)

More possible intel arch strings.

---

Revision: f7969146a5e459cd
Author: Davis King
Date: Mar 09, 2017 (15:14:42 UTC)

Fixed cmake file to correctly enable C++11 by default.

---

Revision: c2286914c854bf5a
Author: Davis King
Date: Mar 07, 2017 (12:19:46 UTC)

Record last changeset and set PATCH version to 99

---

Revision: 74c4985dfb28f1b9
Author: Davis King
Date: Mar 07, 2017 (12:18:27 UTC)

Created release v19.4

---

Revision: bbd3be9b7b0a93f8
Author: Davis King
Date: Mar 05, 2017 (10:36:26 UTC)

Fixed a slow memory leak that occurred when using cuDNN and tensors were created.

---

Revision: d4aef27f3c1e3273
Author: Davis King
Date: Mar 04, 2017 (02:58:55 UTC)

Upped imglab version number

---

Revision: 2f4215c6f8071fb6
Author: Davis King
Date: Mar 04, 2017 (02:58:40 UTC)

Made the image_display widget draw mouse cross-hairs when the user holds shift.

---

Revision: 90c442908635c04c
Author: Davis King
Date: Mar 03, 2017 (11:03:27 UTC)

Added missing const

---

Revision: 03a898e940780408
Author: Davis King
Date: Mar 02, 2017 (00:33:56 UTC)

made test slightly more robust

---

Revision: b368c8702035a874
Author: Davis King
Date: Mar 01, 2017 (04:17:35 UTC)

Set this because CMake search for python is wacked on ubuntu and won't look for new python otherwise.

---

Revision: 647e6aa8a63c8cfe
Author: Davis King
Date: Mar 01, 2017 (00:26:22 UTC)

clarified documentation

---

Revision: d41c70ccd4ed2c8e
Author: Davis King
Date: Feb 28, 2017 (01:01:51 UTC)

Don't give -std=c++11 to nvcc when compiling with visual studio. This avoids a compiler warning.

---

Revision: 56b12b484030b01c
Author: Davis King
Date: Feb 28, 2017 (00:05:25 UTC)

Switched cmake link names to dlib::dlib

---

Revision: fa4e89353d66184b
Author: Davis King
Date: Feb 28, 2017 (-01:43:32 UTC)

Made example cmake use dlib::dlib link name. Also made the installed cmake config for dlib enable C++11.

---

Revision: 6d55f8213d202cda
Author: Davis King
Date: Feb 27, 2017 (14:02:39 UTC)

removed unneeded cruft

---

Revision: 81746022a7a46f7a
Author: Davis King
Date: Feb 27, 2017 (13:44:30 UTC)

Fixed a few cmake errors, one regarding CUDA compilation and another about the all/source.cpp test compile.

---

Revision: 5e058a6abc25cf71
Author: Alain Vaucher
Date: Feb 27, 2017 (19:45:05 UTC)

Delete function that has now become unnecessary

---

Revision: 4e47077557e3c99b
Author: Alain Vaucher
Date: Feb 27, 2017 (19:23:28 UTC)

Change minimum CMake version that supports the introduced features

---

Revision: 0c8bb36a6cbb1a11
Author: Alain Vaucher
Date: Feb 27, 2017 (19:13:44 UTC)

Simplifying file 'cmake' now that include directories are set automatically

---

Revision: 37139fcf88454d1e
Author: Alain Vaucher
Date: Feb 27, 2017 (19:09:40 UTC)

Only specify include directories for dlib_shared if it is built

---

Revision: 157aeb69df697a90
Author: Alain Vaucher
Date: Feb 27, 2017 (19:01:22 UTC)

Add target_include_directories for build interface and for optional libraries

---

Revision: 325fe9637e25d3a3
Author: Alain Vaucher
Date: Feb 27, 2017 (18:48:51 UTC)

Set the include directories automatically for targets that depend on imported dlib

---

Revision: 5f998dbe519f10e7
Author: Evgeniy Fominov
Date: Feb 27, 2017 (17:34:10 UTC)

Allow MINGW OpenBLAS usage (#455)

---

Revision: f6bca6912868edee
Author: Davis King
Date: Feb 25, 2017 (01:42:35 UTC)

clarified spec

---

Revision: 3994ddf2d51dbe51
Author: Ehsan Azar
Date: Feb 25, 2017 (-05:32:07 UTC)

add set_state to kalman filter for the case we know the initial state (#246)

---

Revision: 4ce0a89935fc95bc
Author: Adam Geitgey
Date: Feb 22, 2017 (02:23:21 UTC)

Remove extra semicolons (#446)

---

Revision: 5e707bcee50e10f2
Author: Davis King
Date: Feb 21, 2017 (17:02:45 UTC)

Record last changeset and set PATCH version to 99

---

Revision: 26cdc89f4795a1f9
Author: Davis King
Date: Feb 21, 2017 (17:01:28 UTC)

Created release v19.3

---

Revision: 35b3f33e4e315fdd
Author: Davis King
Date: Feb 20, 2017 (15:45:36 UTC)

clarified spec

---

Revision: 1b1e79fce55c15f9
Author: Davis King
Date: Feb 18, 2017 (18:23:16 UTC)

Made tests more robust and accurate

---

Revision: e097067e9abc21a1
Author: Davis King
Date: Feb 18, 2017 (14:48:58 UTC)

Changed cmake script to avoid warnings from visual studio.

---

Revision: 68afcc5bf7136fb2
Author: Davis King
Date: Feb 18, 2017 (14:20:59 UTC)

Fixed syntax error in cmake script.

---

Revision: fd58d09f23498b8d
Author: Davis King
Date: Feb 18, 2017 (10:33:09 UTC)

Don't build face recognition example if GUI support isn't enabled.

---

Revision: 34c32a7967e36d51
Author: Davis King
Date: Feb 16, 2017 (02:35:43 UTC)

Fixed error in find_gap_between_convex_hulls() test

---

Revision: 6b842e23f51904ea
Author: Davis King
Date: Feb 15, 2017 (16:31:33 UTC)

Added some comments about vector calculus.

---

Revision: 11be214c3dc7736f
Author: Davis King
Date: Feb 15, 2017 (14:41:14 UTC)

merged

---

Revision: 6016a47dbe6d5b2d
Author: Kyle McDonald
Date: Feb 13, 2017 (08:03:37 UTC)

remove anaconda from X11_INCLUDE_DIR. closes #437 (#438)

---

Revision: 3796069f8aa8b5bb
Author: Davis King
Date: Feb 15, 2017 (12:06:03 UTC)

Added find_gap_between_convex_hulls()

---

Revision: ead522d287a8510f
Author: Davis King
Date: Feb 12, 2017 (15:53:02 UTC)

cleanup

---

Revision: 860e151c61ac017c
Author: Davis King
Date: Feb 12, 2017 (15:51:50 UTC)

Fixed silly error in code.

---

Revision: 8aeefae78e1ecd5d
Author: Davis King
Date: Feb 12, 2017 (15:38:17 UTC)

Added a python version of the DNN face recognition example program.

---

Revision: 3709f82c1aac12a9
Author: Davis King
Date: Feb 12, 2017 (15:37:18 UTC)

Added a python interface to the face recognition DNN model.

---

Revision: 9d2a789665221e5f
Author: Davis King
Date: Feb 12, 2017 (11:47:49 UTC)

Make sure fPIC gets propagated to nvcc in all cases.

---

Revision: e6d174d1eb01d2fd
Author: Davis King
Date: Feb 12, 2017 (11:03:40 UTC)

Made full_object_detection comparable via operator==

---

Revision: 77ba01f6b78eaeba
Author: Davis King
Date: Feb 12, 2017 (08:28:42 UTC)

Avoid crashing visual studio

---

Revision: e00f763861885c02
Author: Davis King
Date: Feb 11, 2017 (12:07:51 UTC)

Disable another visual studio warning.

---

Revision: 0225d8a21de6b546
Author: Davis King
Date: Feb 11, 2017 (11:45:22 UTC)

Added try/catch block to main

---

Revision: 79f97bbf6fdf9d50
Author: Davis King
Date: Feb 11, 2017 (10:36:24 UTC)

fixed grammar in comment

---

Revision: fc4e811c9def5e51
Author: Davis King
Date: Feb 11, 2017 (10:24:00 UTC)

clarified metric learning loss

---

Revision: 820d899477e298c4
Author: Davis King
Date: Feb 11, 2017 (09:52:47 UTC)

Added face recognition example

---

Revision: 06845769b55af603
Author: Davis King
Date: Feb 11, 2017 (09:47:44 UTC)

Cleaned up loss_metric_ code a little

---

Revision: 3984c9a36e8f0429
Author: Davis King
Date: Feb 11, 2017 (07:06:46 UTC)

Added roc_point and compute_roc_curve().

---

Revision: b6b46aeecc8de4f5
Author: Davis King
Date: Feb 10, 2017 (15:35:46 UTC)

Fixed incorrect comment.

---

Revision: 86642a2d6d828ff1
Author: Davis King
Date: Feb 09, 2017 (07:38:39 UTC)

Minor cleanup

---

Revision: 14fda140fac5b6a7
Author: Davis King
Date: Feb 08, 2017 (17:53:55 UTC)

Made network smaller.

---

Revision: 8762abd86425a327
Author: Dennis Francis
Date: Feb 07, 2017 (08:52:39 UTC)

feature_addition : Mean squared loss layer for multiple output (#404) (#427) * feature_addition : Mean squared loss layer for multiple output (#404) * Added loss_mean_squared_multioutput layer to support multiple outputs. * Also added a corresponding test case to test a single variable regression with multiple outputs. * Added error checks on truth argument Added assert statements to check that truth argument in compute_loss_value_and_gradient() method contains matrices of correct dimension relative to the output tensor's size. Also the requirements on argument truth to the abstract documentation.

---

Revision: f56ee495ec1e1817
Author: David Miller
Date: Feb 01, 2017 (10:58:40 UTC)

GCC/Clang compatible SIMD code ./dlib/simd/simd**_vec.h (#414) * GCC/Clang compatible vector extension SIMD code * Minimal modifications to dlib for the simd_vec code to work, a few include changes and ifdefs * Changed tabbing to spaces * Allow type inference to binary ops on different types of same size * Added cmake option USE_AUTO_VECTOR, and fixed up preprocessor checks. It is required to build with gcc/clang auto vectorization * Changed to intrinsic version due to poor auto vectorization results. The simd8*_vec are just copies of the C code right now. * Removed _vec variants, added to existing defines. simd_check.h back in place and removed from dlib/simd.h

---

Revision: 4952b019167b9caa
Author: Davis King
Date: Jan 30, 2017 (10:18:39 UTC)

Made alias_tensor usable when it is const.

---

Revision: 73d5520ae19998ab
Author: Davis King
Date: Jan 30, 2017 (06:03:31 UTC)

Made dlib::thread_pool use std::thread and join on the threads in thread_pool's destructor. The previous implementation used dlib's global thread pooling to allocate threads to dlib::thread_pool, however, this sometimes caused annoying behavior when used as part of a MATLAB mex file.

---

Revision: e02b23ca0750aa25
Author: Davis King
Date: Jan 30, 2017 (05:56:47 UTC)

Made global thread pool shutdown not attempted unless the thread pool was already enabled. This also makes the global destruction order a little more deterministic.

---

Revision: 9ed49aea6d41b247
Author: Davis King
Date: Jan 30, 2017 (04:56:06 UTC)

Made matrix literal assignment faster for column major matrices.

---

Revision: fca139c1d34b3c6e
Author: Davis King
Date: Jan 25, 2017 (01:59:08 UTC)

Fixed documentation

---

Revision: 56720f3015d727b2
Author: Daniel Crispell
Date: Jan 24, 2017 (14:11:30 UTC)

add support for loading of RGBA JPEG images (#409)

---

Revision: 27e0fc69d919bf54
Author: Davis King
Date: Jan 22, 2017 (06:32:27 UTC)

merged

---

Revision: 7aa428cdfb3edb2d
Author: RangerUFO
Date: Jan 22, 2017 (31:36:40 UTC)

Fix comments typo in dnn_introduction2_ex (#405)

---

Revision: c17082f59fe25809
Author: Frank
Date: Jan 16, 2017 (06:47:06 UTC)

Fix imglab readme typo in cd path (#400)

---

Revision: 92594614f3de742b
Author: Davis King
Date: Jan 22, 2017 (05:25:29 UTC)

Added comments about test_one_step()

---

Revision: df6d88905db8d270
Author: Davis King
Date: Jan 22, 2017 (05:25:06 UTC)

Added test_one_step() to the dnn_trainer. This allows you to do automatic early stopping based on observing the loss on held out data.

---

Revision: 13935a74049b696b
Author: Davis King
Date: Jan 21, 2017 (16:26:15 UTC)

Always compile dnn code.

---

Revision: c80e888bdac79aa4
Author: Stefan Schweter
Date: Jan 08, 2017 (21:46:35 UTC)

Made matrix2 test more robust on ARM64 (#394)

---

Revision: 8cba3cbdd468dbc8
Author: Davis King
Date: Jan 09, 2017 (02:14:20 UTC)

Made sure cmake config file pulls in dependent include folders.

---

Revision: 7be4ff22448d3a93
Author: Davis King
Date: Jan 06, 2017 (11:12:54 UTC)

Only turn off CUDA_PROPAGATE_HOST_FLAGS on OS X since that's the only place that really needs it off.

---

Revision: 450b23e56c3a9f10
Author: ernestotapiar
Date: Jan 06, 2017 (04:04:41 UTC)

cmake script to suppress randlib warnings about *.a having no symbols on MacOSX (#390) * cmake script to suppress randlib warnings about *.a having no symbols on MacOSX * Moved script code into the main CMakeLists.txt file to suppress the superfluous ranlib warnings all the time.

---

Revision: b28ea90fa729da6c
Author: Davis King
Date: Jan 04, 2017 (12:19:31 UTC)

Changed to avoid linker errors on some compilers.

---

Revision: 17c7c2219342a49a
Author: Davis King
Date: Jan 03, 2017 (01:35:32 UTC)

Initialized exception_ptr since some systems, incorrectly, don't initialize it automatically.

---

Revision: 253dfed88be54577
Author: rquk
Date: Jan 02, 2017 (18:08:03 UTC)

Fixed issue with uninitialised variables in thread_pool_extension. (#386) * Fixed issue with uninitialised variables. There are 2 places where std::exception_ptr eptr is not initialised. * running_gradient.h needs to qualify erfc with 'std::' to avoid compilation error. Found when compiling dnn_mmod_face_detection_ex.cpp with RadStudio and Clang compiler.

---

Revision: f88e2a0430a03561
Author: Davis King
Date: Jan 04, 2017 (04:10:13 UTC)

Fixed cmake script so that nvcc gets the correct compiler options sent to it (in particular on mac os x, where this flag needs to be set this way for it to work in some cases).

---

Revision: ff61b258c75b289c
Author: Davis King
Date: Dec 27, 2016 (04:33:41 UTC)

Made pixel_traits use constexpr everywhere to avoid compiler errors in newer compilers.

---

Revision: a340e4781e58c6e1
Author: Stefan Schweter
Date: Dec 24, 2016 (03:09:12 UTC)

Added new test case for split method (unicode non-breaking space) (#374)

---

Revision: 2e7eb3789277ac0a
Author: Davis King
Date: Dec 26, 2016 (03:56:40 UTC)

Made the dlib shared library build dependent on the static library build. This is because some build systems seem to mess up the parallel build of these two things and sometimes rebuild and potentially clobber object files during the parallel build. I haven't seen this result in a problem, but the race condition does seem to exist in some build systems. So this should force the two top level targets to be build sequentially rather than in parallel, thus avoiding any possible race condition.

---

Revision: a0f414b612966b53
Author: Davis King
Date: Dec 25, 2016 (07:32:48 UTC)

fixed spelling error in comment

---

Revision: 457530e2888669e8
Author: Davis King
Date: Dec 23, 2016 (05:12:12 UTC)

Added matlab_object to the mex wrapper. Now you can have parameters that are arbitrary matlab objects.

---

Revision: 1ea53fefe15fbd03
Author: Davis King
Date: Dec 22, 2016 (15:43:50 UTC)

Removed C++11 check here since it's now redundant with the checks in algs.h. Also, the check doesn't work correctly in visual studio since visual studio doesn't define __cplusplus to 201103 even though the current version ostensibly supports C++11.

---

Revision: f003087ebdd79225
Author: Juha Reunanen
Date: Dec 20, 2016 (15:35:13 UTC)

Minor documentation fixes (#368)

---

Revision: 917332ad117dac01
Author: Davis E. King
Date: Dec 19, 2016 (10:14:49 UTC)

Merge pull request #365 from reunanen/fix-363 Fix #363 by not setting CUDA_HOST_COMPILER when targeting Visual Studio

---

Revision: cc6fec145c693d78
Author: Juha Reunanen
Date: Dec 19, 2016 (21:02:27 UTC)

Set CUDA_HOST_COMPILER when not targeting Visual Studio

---

Revision: fdf84afc2cc22701
Author: Juha Reunanen
Date: Dec 19, 2016 (18:10:38 UTC)

Workaround for #363: do not set CUDA_HOST_COMPILER for the CUDA test

---

Revision: fe39427ad616dd3a
Author: Juha Reunanen
Date: Dec 19, 2016 (19:07:43 UTC)

Problem: Visual Studio linker errors resulting from mixed /MT and /MD settings Solution: don't set CUDA_PROPAGATE_HOST_FLAGS to OFF when using Visual Studio

---

Revision: 5821cafa67ecc8f6
Author: Davis King
Date: Dec 18, 2016 (18:35:25 UTC)

Changed build order to see if it prevents travis from crashing.

---

Revision: 0d9340beee693521
Author: Davis King
Date: Dec 18, 2016 (15:52:45 UTC)

Minor improvement to example

---

Revision: 2088fe75c0c744af
Author: Davis King
Date: Dec 18, 2016 (08:20:37 UTC)

more cleanup

---

Revision: 1c3bc4eec6e9e184
Author: Davis King
Date: Dec 18, 2016 (08:10:13 UTC)

Improved example

---

Revision: 7e33779c306dec13
Author: Davis King
Date: Dec 17, 2016 (15:41:36 UTC)

Finished the more complex metric learning example and added some example data.

---

Revision: 70cd167f83ef7711
Author: Davis King
Date: Dec 17, 2016 (11:46:39 UTC)

Improved example

---

Revision: 3254c571384e1834
Author: Davis King
Date: Dec 17, 2016 (11:11:18 UTC)

Cleaned up example

---

Revision: 96e3596b9d8c79a0
Author: Davis King
Date: Dec 17, 2016 (10:08:58 UTC)

fixed grammar

---

Revision: b2164124cf92daa0
Author: Davis King
Date: Dec 17, 2016 (09:29:29 UTC)

Added another metric learning example

---

Revision: 6986f126dc8e8e6d
Author: Davis King
Date: Dec 17, 2016 (08:37:01 UTC)

Made get_net() sync to disk if the disk date is out of date. This way, when using train_one_step(), you will get the behavior of automatic disk syncs at the end of training.

---

Revision: 5d5249551797d05f
Author: Davis King
Date: Dec 17, 2016 (05:56:27 UTC)

Added metric learning example

---

Revision: bfe738d443233a60
Author: Davis King
Date: Dec 17, 2016 (05:45:55 UTC)

clarified spec

---

Revision: 0559cc5e7bbda8e3
Author: Davis King
Date: Dec 17, 2016 (05:30:46 UTC)

Added docs and cleaned up code slightly.

---

Revision: ee835f2b30370b1f
Author: Davis King
Date: Dec 17, 2016 (04:45:03 UTC)

Removed loss_metric_hardish_

---

Revision: 25c22f567aebab2e
Author: Davis King
Date: Dec 15, 2016 (11:06:37 UTC)

Allow enqueueing to a pipe from rvalues.

---

Revision: d6a6124156e4293b
Author: Davis King
Date: Dec 15, 2016 (04:07:04 UTC)

Fixed incorrect mex file name being shown in the generated class code.

---

Revision: f2d66cf532a73125
Author: Davis King
Date: Dec 14, 2016 (10:09:53 UTC)

Improved how cmake sets up clang and cuda. This is basically to work around partial C++11 support and funny cmake behavior of clang on OS X.

---

Revision: 7c9b777fbd84469a
Author: Davis King
Date: Dec 14, 2016 (10:08:45 UTC)

Made this test try to enable C++11 via the compiler flag.

---

Revision: 3ad63d456a997bb5
Author: Davis King
Date: Dec 14, 2016 (04:51:03 UTC)

Updated C++11 enabling script to work around some kind of bug in cmake related to new versions of clang on OSX, without this extra statement C++11 wasn't being enabled.

---

Revision: e91efa76ed9645ba
Author: Davis King
Date: Dec 13, 2016 (04:57:07 UTC)

Suppressed cmake warning

---

Revision: e6021de44315bd86
Author: Davis King
Date: Dec 13, 2016 (04:36:55 UTC)

Set set(CMAKE_CXX_STANDARD_REQUIRED YES)

---

Revision: f975dd4cc28ce392
Author: Davis King
Date: Dec 10, 2016 (06:20:20 UTC)

Relaxed requires clauses even more.

---

Revision: 8e794a4fd291d677
Author: Davis King
Date: Dec 10, 2016 (06:03:07 UTC)

Relaxed require clause

---

Revision: 51c6afcaa41c826d
Author: Davis King
Date: Dec 10, 2016 (03:36:45 UTC)

Added an overload of flip_image_left_right() that operates inplace.

---

Revision: e47f00a5291e31ee
Author: Davis King
Date: Dec 09, 2016 (16:05:18 UTC)

Relaxed requires clause slightly

---

Revision: b71b824c942b4de2
Author: Davis King
Date: Dec 09, 2016 (14:51:58 UTC)

Made dlib::array movable.

---

Revision: 13b679aea0be116b
Author: Davis King
Date: Nov 30, 2016 (05:36:32 UTC)

Fixed missing implementation of map_input_to_output() and map_output_to_input() in the concat_ layer.

---

Revision: 7506978fb45dc29a
Author: Davis King
Date: Nov 29, 2016 (06:05:16 UTC)

Added running_stats_decayed

---

Revision: 0c40eff84c399624
Author: Davis King
Date: Nov 28, 2016 (15:03:17 UTC)

Made dlib/cmake add the appropriate folders to client include search paths.

---

Revision: 475a03e598fb12a3
Author: Duncan Palmer
Date: Nov 28, 2016 (26:07:14 UTC)

Ensure MKL detection doesn't depend on MKL header files being present. Apparently, MKL installations don't always include header files. So detect MKL libraries and headers seperately.

---

Revision: f52ad8692907a4b6
Author: Duncan Palmer
Date: Nov 23, 2016 (14:05:13 UTC)

Seperate MKL FFT and MKL BLAS configurations. MKL FFT bindings had assumed that MKL was also being used to provide the BLAS implementation, however the build system allows these 2 things to be configured independantly.

---

Revision: 685838d8af7b5b94
Author: Duncan Palmer
Date: Sep 08, 2016 (24:41:52 UTC)

Add MKL DFTI FFT bindings.

---

Revision: a9593a0555281389
Author: Davis King
Date: Nov 27, 2016 (08:24:16 UTC)

clarified spec

---

Revision: 555f7f98a3321165
Author: Davis King
Date: Nov 27, 2016 (08:20:10 UTC)

Added segment_number_line().

---

Revision: 84dfcad68385e915
Author: Davis King
Date: Nov 26, 2016 (10:48:00 UTC)

Added append() to random_cropper.

---

Revision: 9510bb9314401fa6
Author: Davis King
Date: Nov 26, 2016 (03:26:59 UTC)

Made the random translation amount user settable.

---

Revision: b0644464754f58e6
Author: Davis King
Date: Nov 26, 2016 (03:17:24 UTC)

Changed to avoid gcc warnings

---

Revision: 6f7b3c6bf20a20b0
Author: Dennis Francis
Date: Nov 26, 2016 (14:37:36 UTC)

adapt to dlib indentation style

---

Revision: 6522d8c5785742a5
Author: Dennis Francis
Date: Nov 25, 2016 (27:45:39 UTC)

converted tabs to spaces in the indentation

---

Revision: a460abd4fd4d0d9f
Author: Dennis Francis
Date: Nov 23, 2016 (19:74:33 UTC)

feature_addition : Added a mean squared loss layer to DNN Added mean squared loss layer "loss_mean_squared" to DNN as requested in https://github.com/davisking/dlib/issues/152 Also added test case of a simple linear regression with one variable that uses this layer.

---

Revision: 7e3cf5042c2a9ead
Author: Stefan Schweter
Date: Nov 19, 2016 (18:54:56 UTC)

Disable misleading-indentation warning in unicode.h (pragma for GCC >= 6)

---

Revision: d17ddfe34f8e7177
Author: Davis King
Date: Nov 25, 2016 (02:46:52 UTC)

Clarified spec

---

Revision: cc5d9caedb5c1105
Author: Davis King
Date: Nov 22, 2016 (07:59:24 UTC)

Fixed compile time error in random_subset_selector::swap()

---

Revision: 29c4d3ea7bd70110
Author: Davis King
Date: Nov 19, 2016 (10:37:51 UTC)

Made the dnn_trainer's detection and backtracking from situations with increasing loss more robust. Now it will never get into a situation where it backtracks over and over. Instead, it will only backtrack a few times in a row before just letting SGD run unimpeded.

---

Revision: 1a86b32362308860
Author: Davis King
Date: Nov 18, 2016 (08:47:50 UTC)

Made tt::add() more general

---

Revision: 4e9fa80c14f028cd
Author: Davis King
Date: Nov 18, 2016 (07:40:01 UTC)

Suppress bogus warnings from nvcc about unreachable code that is actually reachable.

---

Revision: 9a400902da27a32c
Author: Davis King
Date: Nov 18, 2016 (05:42:03 UTC)

Added min_pointwise() and max_pointwise().

---

Revision: 0f33099b2f19a5ce
Author: Davis King
Date: Nov 17, 2016 (07:10:54 UTC)

Made non-const alias tensors convertible to const alias tensors.

---

Revision: 358dc20ad776fd4a
Author: Davis King
Date: Nov 15, 2016 (01:56:00 UTC)

Added missing #include

---

Revision: c9a9b5c662feb7e0
Author: caseymcc
Date: Nov 14, 2016 (07:52:45 UTC)

Remove initializer lists support for Apple/Clang (libstdc++ version used does not include std::initializer_list)

---

Revision: df3a4de43efb9f0c
Author: Davis King
Date: Nov 14, 2016 (10:47:04 UTC)

Avoid division by zero in running_scalar_covariance_decayed.

---

Revision: 403deb128098f6ae
Author: Davis King
Date: Nov 14, 2016 (06:06:56 UTC)

Added running_scalar_covariance_decayd

---

Revision: de112e9f01f1b126
Author: Avi Haiat
Date: Nov 12, 2016 (16:54:48 UTC)

Expose python binding for running multiple detectors at once (#328) * Expose python binding for running multiple detectors at once * Remove unwanted typedef

---

Revision: ae4e3984e1d04ccc
Author: MambaWong
Date: Nov 12, 2016 (19:39:05 UTC)

Modify to support both train-images-idx3-ubyte and train-images.idx3-ubyte styles of the filename. (#327)

---

Revision: dbc4a0c8365e10ce
Author: Avi Haiat
Date: Nov 10, 2016 (05:36:52 UTC)

Add option --rm-other-labels to imglab (#323) * Add option --rm-other-labels to imglab * Add support for comma separated list of labels

---

Revision: 44c0e160fac405c1
Author: Davis King
Date: Nov 09, 2016 (05:30:16 UTC)

Fixed rls numerical instability.

---

Revision: 68b8859404b273dd
Author: Davis King
Date: Nov 08, 2016 (09:40:19 UTC)

Made rls run a bit faster, especially if the new mode that allows the regularization to decay away is activated.

---

Revision: 151810f94cabb586
Author: Davis King
Date: Nov 08, 2016 (06:03:40 UTC)

Fixed matlab crashing when an error message from a mex file included the % character, since that is interpreted by matlab as part of an eventual printf() code.

---

Revision: 69ac1166bbd80f50
Author: Davis King
Date: Nov 07, 2016 (14:50:12 UTC)

Make sure the state of the assert macros is always consistent between how dlib is built as a stand alone library and what appears in the config.h file.

---

Revision: 54d0970d0a953330
Author: Davis King
Date: Nov 06, 2016 (08:53:22 UTC)

Adding more possible metric loss layers.

---

Revision: 37b2375391469e62
Author: Davis King
Date: Nov 05, 2016 (17:12:33 UTC)

Changed the loss layer interface to use two typedefs, output_label_type and training_label_type instead of a single label_type. This way, the label type used for training can be distinct from the type output by the network. This change breaks backwards compatibility with the previous API.

---

Revision: aa3bbd123ccdcb09
Author: Davis King
Date: Nov 05, 2016 (06:52:09 UTC)

Added serialization support for std::array.

---

Revision: d7710f6301efea99
Author: Davis King
Date: Nov 04, 2016 (16:30:03 UTC)

Changed CMake so it doesn't automatically enable asserts in debug builds since this frustrates Visual Studio users who want to compile dlib as a static library.

---

Revision: b212277ed1a04acd
Author: Stefan Schweter
Date: Nov 04, 2016 (22:55:25 UTC)

Explicitly use signed char in max_cost_assignment test (fix for ARM/ARM64) (#316)

---

Revision: ed8309665c022d71
Author: Stefan Schweter
Date: Nov 04, 2016 (19:48:09 UTC)

Made active_learning test more robust on ARM64. (#315)

---

Revision: 9e3cb4d950630b72
Author: caseymcc
Date: Oct 28, 2016 (09:15:55 UTC)

fix include issue when using libjpeg compiled outside of dlib (#307)

---

Revision: da705c1bfa258481
Author: Davis King
Date: Nov 02, 2016 (06:12:07 UTC)

Fixed compile time error.

---

Revision: 0f44bd64a3b6da0e
Author: Davis King
Date: Nov 02, 2016 (06:06:36 UTC)

Wrote replacements for set_tensor() and scale_tensor() since the previous versions were calling into cuDNN, however, the cuDNN functions for doing this are horrifically slow, well over 100x slower than they should be, which is surprising since these functions are so trivial.

---

Revision: 40de2fbd2bfbf0a2
Author: Davis King
Date: Oct 30, 2016 (03:42:27 UTC)

Made test deterministic

---

Revision: 6fd6fcb3ae8f4383
Author: Davis King
Date: Oct 27, 2016 (07:29:27 UTC)

Fixed some race conditions in cuda code.

---

Revision: 3fb9cb42908da65e
Author: Davis King
Date: Oct 26, 2016 (13:09:44 UTC)

Fixed compiler warnings

---

Revision: f6597d42842009ad
Author: Davis King
Date: Oct 26, 2016 (13:06:02 UTC)

Added tt::scale_columns(). Also optimized some other cuda kernel launches a little.

---

Revision: cf213dfacd1ebded
Author: Davis King
Date: Oct 26, 2016 (12:49:56 UTC)

Added is_vector() for tensor objects.

---

Revision: 7ecdc2e44d57a754
Author: Davis King
Date: Oct 26, 2016 (05:06:05 UTC)

Improved cudnn error messages.

---

Revision: 5e6cf2433e49ba1d
Author: Davis King
Date: Oct 23, 2016 (11:13:46 UTC)

Added l2normalize_ layer.

---

Revision: 2076132a771b0907
Author: Davis King
Date: Oct 23, 2016 (06:44:05 UTC)

Made approximate derivative calculation slightly more numerically robust.

---

Revision: 36d2e4b214b33488
Author: Davis King
Date: Oct 21, 2016 (18:48:03 UTC)

Made the dnn_trainer automatically reload from the last good state if a loss of NaN is encountered.

---

Revision: eaf32ef301e22fcd
Author: Davis King
Date: Oct 18, 2016 (18:54:34 UTC)

Fixed a bug in the repeat layer that caused it to throw exceptions in some cases.

---

Revision: 3003fe9808305f60
Author: Davis King
Date: Oct 16, 2016 (15:51:19 UTC)

Added --rmignore

---

Revision: 007a4fabe1cfd6ac
Author: Davis King
Date: Oct 16, 2016 (12:38:17 UTC)

Added loss_metric_

---

Revision: 6ecb13bd4efdc2f4
Author: Davis King
Date: Oct 16, 2016 (12:21:28 UTC)

Improved xml logging for --resample

---

Revision: 9dc1723d0ec278d0
Author: Davis King
Date: Oct 16, 2016 (12:19:24 UTC)

removed --extract-chips

---

Revision: 969b13255a40430a
Author: Davis King
Date: Oct 16, 2016 (12:10:44 UTC)

Fixed confusing -h docs

---

Revision: 91364e2f189cb037
Author: Davis King
Date: Oct 16, 2016 (12:09:06 UTC)

Added --jpg and --one-object-per-image.

---

Revision: 3cb6c541e2a9e01d
Author: Evgeniy Fominov
Date: Oct 14, 2016 (24:44:31 UTC)

Travis CI Integration scripts (#294)

---

Revision: d8b1a27bf2732380
Author: hannometer
Date: Oct 14, 2016 (16:35:09 UTC)

pass CUDA_HOST_COMPILER variable to cuda cmake tests (#292)

---

Revision: b7900c78a0328bbc
Author: Mateusz Baran
Date: Oct 14, 2016 (14:54:59 UTC)

optimization - poly_min_extrap fixed for a specific case (#291) If temp is nonzero but a subnormal then temp2 matrix may contain infinities which may cause temp assignment in line 225 to be a NaN (as a difference of two positive infinities).

---

Revision: 0e231bb9d3f2da6a
Author: Davis King
Date: Oct 10, 2016 (15:36:19 UTC)

Record last changeset and set PATCH version to 99

---

Revision: f8fa027c760270d8
Author: Davis King
Date: Oct 10, 2016 (15:34:32 UTC)

Created release v19.2

---

Revision: 9d2cc9605dabbc51
Author: Davis King
Date: Oct 10, 2016 (14:26:09 UTC)

Made tensor copies a little more efficient.

---

Revision: c6c9fdeaa6415cde
Author: Evgeniy Fominov
Date: Oct 10, 2016 (18:32:16 UTC)

Added threads.h into shape_predictor.h (#283)

---

Revision: a37ad2dbe73ebd80
Author: Evgeniy Fominov
Date: Oct 10, 2016 (18:30:22 UTC)

Fix for #281 (#282)

---

Revision: 17e2aad4fd065e39
Author: Davis King
Date: Oct 10, 2016 (04:12:58 UTC)

Added missing #include.

---

Revision: 3e63dfdda30dc7e3
Author: Davis King
Date: Oct 10, 2016 (03:36:10 UTC)

Added missing #include

---

Revision: 38e96e999d2cebb3
Author: Davis King
Date: Oct 09, 2016 (19:29:20 UTC)

Removing mltool since I'm pretty sure no one uses it and it might even cause user confusion.

---

Revision: 0276c3e7b67d2eef
Author: Davis King
Date: Oct 09, 2016 (19:18:46 UTC)

Made unit tests compile stuff in tools folder

---

Revision: 96a3a68254e12408
Author: Davis King
Date: Oct 09, 2016 (19:18:12 UTC)

Minor code cleanup.

---

Revision: be4f95c3025c0c8a
Author: Davis King
Date: Oct 09, 2016 (14:25:52 UTC)

Improved check for old visual studio

---

Revision: f1e84240b9f944fd
Author: Davis King
Date: Oct 09, 2016 (14:19:51 UTC)

Fixed old visual studio check

---

Revision: 3f65c6abd14c2ffa
Author: Davis King
Date: Oct 09, 2016 (14:13:21 UTC)

Improved error messages about non-C++11 compilers and made the example cmake file a little clearer on this issue.

---

Revision: 0cb043aab1a94228
Author: Davis King
Date: Oct 09, 2016 (12:36:48 UTC)

Renamed file so it matches the naming scheme of the other matlab examples.

---

Revision: 0a5f86061b25eeb4
Author: Davis King
Date: Oct 09, 2016 (06:38:03 UTC)

Moved all the test_object_detection_function() specs into one file so they are organized in the HTML documentation.

---

Revision: b87b1987c2fa72de
Author: Davis King
Date: Oct 08, 2016 (11:28:58 UTC)

Changed cmake so it compiles most of the DNN examples when using visual studio.

---

Revision: 5568c09ce2ef01e1
Author: Davis King
Date: Oct 08, 2016 (10:28:21 UTC)

Added warning message about trying to use cuda from a 32 bit visual studio project.

---

Revision: 09538bd6884d5d85
Author: Davis King
Date: Oct 08, 2016 (10:27:34 UTC)

Evgeniy Fominov's changes that avoid compiler bugs in Visual Studio.

---

Revision: 1dbb13122991d995
Author: Davis King
Date: Oct 08, 2016 (10:26:33 UTC)

Minor change to avoid warnings from visual studio.

---

Revision: 1f0c4fd8dc47fb83
Author: Davis King
Date: Oct 07, 2016 (18:07:13 UTC)

Made the upsampling conditional on the image not being huge already.

---

Revision: 2a834745327ca5c7
Author: elda27
Date: Oct 07, 2016 (28:36:58 UTC)

find_cudnn modified for cudnn installed a user directory. (#275)

---

Revision: 9251f7002f1f5525
Author: Jan Rüegg
Date: Oct 05, 2016 (19:15:31 UTC)

Unreachable code, fixed compiler warning (#272) * Unreachable code, fixed compiler warning * Update simd4f.h

---

Revision: fa8285990c12208b
Author: Jan Rüegg
Date: Oct 05, 2016 (18:57:01 UTC)

Unreachable code, fixed compiler warning (#271) * Unreachable code, fixed compiler warning * Update simd8f.h

---

Revision: 27148e269324a6a7
Author: Davis King
Date: Oct 05, 2016 (06:23:02 UTC)

Made the dnn_trainer check if the loss has been increasing before it saves the state to disk. If it detects that the loss has been going up then instead of saving to disk it recalls the previously good state. This way, if we hit a bad mini-batch during training which negatively effects the model in a significant way, the dnn_trainer will automatically revert back to an earlier good state.

---

Revision: 6ae4620f2ac43b69
Author: Davis King
Date: Oct 03, 2016 (15:55:31 UTC)

Made find_upper_quantile() more general

---

Revision: 4333eb792b01417e
Author: Davis King
Date: Oct 02, 2016 (13:52:39 UTC)

Merge branch 'skip_gui_examples' of git://github.com/miscellanea/dlib into miscellanea-skip_gui_examples Conflicts: examples/CMakeLists.txt

---

Revision: 0c4e56915a1f6f28
Author: Xi Deng
Date: Oct 01, 2016 (11:49:57 UTC)

skip examples which need GUI when DLIB_NO_GUI_SUPPORT is defined

---

Revision: 56a8a6b0e3c876be
Author: Davis King
Date: Oct 02, 2016 (12:43:11 UTC)

Added comments

---

Revision: 35746eebbe5ae72e
Author: Davis King
Date: Oct 02, 2016 (10:33:22 UTC)

Larger mustache :)

---

Revision: 796613c025c742ff
Author: Davis King
Date: Oct 02, 2016 (09:37:47 UTC)

Added test image for dog mmod example program

---

Revision: 1a1ea5f589d10e55
Author: Davis King
Date: Oct 02, 2016 (09:00:07 UTC)

Added more mmod examples.

---

Revision: 4ecbc95dd20e02b2
Author: Davis King
Date: Oct 02, 2016 (08:54:14 UTC)

Simplified the network definition slightly

---

Revision: 832dd2b780d5dcc9
Author: Davis King
Date: Oct 02, 2016 (05:25:19 UTC)

Added image_dataset_file::shrink_big_images(). So now load_image_dataset() can load a dataset of high resolution files into a user requested lower resolution.

---

Revision: d1edff75d500b46e
Author: miscellanea
Date: Oct 01, 2016 (10:52:45 UTC)

Add arm files for libpng from official libpng v1.6.7 to support neon (#262) (#264) Now dlib can be compiled for android with -DANDROID_ABI="armeabi-v7a with NEON" with https://github.com/taka-no-me/android-cmake However, when compiling for arm64-v8a, since __ARM_NEON__ is not defined, neon code will not be actually enabled, which requires upgrading libpng a bit more

---

Revision: ae14992c6332d828
Author: Davis King
Date: Oct 01, 2016 (14:33:49 UTC)

Made the check in dnn_trainer for convergence more robust. Previously, if we encountered a bad mini-batch that made the loss value suddenly jump up by a larger than normal value it could make the trainer think we converged. Now the test is robust to recent spikes in loss value.

---

Revision: 745de03984577150
Author: Davis King
Date: Oct 01, 2016 (14:32:01 UTC)

Added find_upper_quantile() and count_steps_without_decrease_robust().

---

Revision: b26b4179004a9a68
Author: Davis King
Date: Oct 01, 2016 (14:00:36 UTC)

Changed DEFAULT_BATCH_NORM_EPS from 1e-5 to 1e-4.

---

Revision: 52b7ca67fde80387
Author: Davis King
Date: Oct 01, 2016 (11:12:07 UTC)

Removed the last bit of code with any heritage from numerical recipes in C. This was in some of the svd routines. However, we already had a svd routine that used a separate svd code that is better than the NRIC derived version. So that's what we use everywhere now.

---

Revision: 7a017c493450a5a3
Author: Davis King
Date: Oct 01, 2016 (09:20:16 UTC)

Minor change to avoid warning from gcc

---

Revision: 5dd0cfca877caab1
Author: Davis King
Date: Sep 30, 2016 (19:09:03 UTC)

Fixed spelling error in comment

---

Revision: 57bd723f7e7d8d75
Author: Davis King
Date: Sep 30, 2016 (17:06:11 UTC)

Minor changes to avoid warnings in visual studio 2015

---

Revision: 16774140d2203404
Author: Davis King
Date: Sep 29, 2016 (04:57:01 UTC)

Fixed spelling error in comment

---

Revision: 8f5f924a938d99ea
Author: Eugene
Date: Sep 29, 2016 (03:15:25 UTC)

fixed warning about unsigned-signed comparison in imgtool (#258)

---

Revision: aae42d24e3e0ae74
Author: Davis King
Date: Sep 28, 2016 (17:14:20 UTC)

Improved error message.

---

Revision: c772c396f47ba1be
Author: Davis King
Date: Sep 28, 2016 (17:13:43 UTC)

Changed default background cropping rate from 0.1 to 0.5.

---

Revision: b84a32d32bde42e2
Author: Davis King
Date: Sep 28, 2016 (17:13:08 UTC)

Added missing requires clauses and asserts.

---

Revision: 8b677387a30aa788
Author: Davis King
Date: Sep 28, 2016 (17:01:49 UTC)

Added --sort-num-objects and cleaned up code slightly.

---

Revision: 208920d77ff90da6
Author: Ehsan Azar
Date: Sep 21, 2016 (09:36:24 UTC)

first check large number in elsif (#239)

---

Revision: afa9ee5ce6d8ce61
Author: Evgeniy Fominov
Date: Sep 21, 2016 (17:19:13 UTC)

Fixed warning about winsock2.h in windows build (#236)

---

Revision: 703e614feade0698
Author: Davis King
Date: Sep 28, 2016 (05:30:17 UTC)

Made cmake automatically increase clang's template depth limit so the dnn imagenet examples compile.

---

Revision: 757e906e1148853c
Author: Davis King
Date: Sep 28, 2016 (05:28:40 UTC)

Fixed --cluster not working with relative paths.

---

Revision: 0cbf2bbfde39e2bf
Author: Davis King
Date: Sep 26, 2016 (16:50:54 UTC)

Added --rmlabel and --rm-if-overlaps. Also changed the behavior of --split so that it simply partitions the data and is an invertible operation.

---

Revision: 1a9cff21810532ab
Author: Davis King
Date: Sep 21, 2016 (03:01:54 UTC)

Made thread_pool's destructor call abort() rather than rethrow any exception to avoid compiler warnings.

---

Revision: 2b44381e5800ba86
Author: Davis King
Date: Sep 18, 2016 (07:54:12 UTC)

Made --cluster skip empty images

---

Revision: b4cc8af1ea94c05a
Author: Davis King
Date: Sep 18, 2016 (05:37:23 UTC)

Improved CLI for --resample

---

Revision: 11c1841ef60b2f62
Author: Davis King
Date: Sep 18, 2016 (05:09:39 UTC)

Fixed a bug in --cluster where it would output xml files with empty entries if the input xml file contained unannotated images.

---

Revision: d84791b084a95bf1
Author: Davis King
Date: Sep 17, 2016 (11:19:52 UTC)

Fixed getting cli warning message when gui closed.

---

Revision: 5a82dbc2bb81da11
Author: Davis King
Date: Sep 17, 2016 (05:12:33 UTC)

Made imglab's --resample not skip objects near the edge of the image.

---

Revision: c41a595d73a7ac90
Author: Davis King
Date: Sep 16, 2016 (15:45:06 UTC)

Added message about -h

---

Revision: dde8f363dbc20329
Author: Davis King
Date: Sep 16, 2016 (02:47:49 UTC)

Made the background crops output by the random_cropper a little more expansive.

---

Revision: 7047f2d34a40e711
Author: Davis King
Date: Sep 16, 2016 (02:46:35 UTC)

Added --rmempty to imglab

---

Revision: 6709dd779319d5a6
Author: Gregor Fabritius
Date: Sep 13, 2016 (04:04:48 UTC)

Fixed cudnn_hint_path detection for OSX where CUDA_CUBLAS_LIBRARIES is a list, not a single string. (#230)

---

Revision: 0632c645f4611695
Author: Davis King
Date: Sep 12, 2016 (16:00:37 UTC)

Made this script more robust

---

Revision: b17024aba9803f5b
Author: Davis King
Date: Sep 11, 2016 (04:54:26 UTC)

Added an error message that prints to STDERR when you accidentally let an exception sit in a thread pool and don't handle it before the thread pool is destroyed.

---

Revision: 8f1b0437dbbf6719
Author: Davis King
Date: Sep 10, 2016 (13:33:32 UTC)

Made the dnn_trainer propagate exceptions that happen during training (in its training thread) out of the object into the calling code rather than terminating the application.

---

Revision: 052c2dad2674d347
Author: Davis King
Date: Sep 10, 2016 (11:29:06 UTC)

Made input_rgb_image_pyramid use multiple cores while building the pyramid.

---

Revision: 7048cc65daaa69b5
Author: Lucas Clemente Vella
Date: Sep 10, 2016 (12:31:01 UTC)

Adding specific parameters interface on fc_ layer (#213)

---

Revision: a10e399cd00a7f13
Author: Davis King
Date: Sep 09, 2016 (16:22:49 UTC)

Fix cmake script so they don't think visual studio 2015 supports C++11, because it doesn't. :(

---

Revision: 3fd7471574689a16
Author: Gregor Fabritius
Date: Sep 09, 2016 (14:48:29 UTC)

Fixed detection of C++11 features (#224) Detecting C++11 capabilities did not work on OSX 10.11.6, XCode 7.3.1, Apple LLVM 7.3.0, cmake 3.6.1. First check revealed that CMAKE_CXX_KNOWN_FEATURES was empty, so subsequent matches failed. After reading http://public.kitware.com/pipermail/cmake/2014-November/059182.html changed handling to fetch property instead of relying on variable. Works on my machine.

---

Revision: 5dedeb669dc8342f
Author: Davis King
Date: Sep 07, 2016 (17:12:01 UTC)

Added --min-object-size option to imglab.

---

Revision: 2ee81fb13a873e43
Author: Davis King
Date: Sep 06, 2016 (07:41:50 UTC)

Added a warning message to remind users to enable C++11 support.

---

Revision: 86b71144e75346ec
Author: Davis King
Date: Sep 06, 2016 (04:15:56 UTC)

Removed compile statements for older versions of visual studio

---

Revision: 127f8e4fb9e076c0
Author: Davis King
Date: Sep 06, 2016 (04:13:55 UTC)

Added compiler error messages indicating that newer versions of dlib no longer work with older versions of Visual Studio. This is because C++11 support is too bad in older versions of Visual Studio.

---

Revision: 25efb06472834588
Author: Davis King
Date: Sep 06, 2016 (03:03:30 UTC)

Added a check for non-finite values to the trust region solver.

---

Revision: e714cc950c2bc117
Author: Davis King
Date: Sep 05, 2016 (11:40:54 UTC)

Fixed example

---

Revision: 82d47f51f8e36f6c
Author: Davis King
Date: Sep 05, 2016 (11:39:31 UTC)

Added random_cropper and DNN MMOD example programs.

---

Revision: 84e726857717fbc7
Author: Davis King
Date: Sep 05, 2016 (11:38:44 UTC)

Adjusted face boxes so they are a little bit more accurately placed.

---

Revision: 8e8d1265bbddfffe
Author: Davis King
Date: Sep 05, 2016 (11:00:52 UTC)

Added a call to set_all_bn_running_stats_window_sizes() to set this value to what it was when this example was trained (since I just modified the default value in dlib to something else).

---

Revision: 7cb6a3b374735c66
Author: Davis King
Date: Sep 05, 2016 (10:55:53 UTC)

Added a method to set the random_cropper's random seed.

---

Revision: 0ab06c47c3704900
Author: Davis King
Date: Sep 05, 2016 (10:48:17 UTC)

Gave the random_cropper the option to randomly include background patches.

---

Revision: 23974ad798bab91e
Author: Davis King
Date: Sep 05, 2016 (06:15:52 UTC)

Added overload of member function set_chip_dims() and also cleaned up a few minor details.

---

Revision: 0131ae15b53907f1
Author: Davis King
Date: Sep 05, 2016 (06:15:22 UTC)

Added another load_image_dataset() overload for string filenames

---

Revision: c946d61199c1faa6
Author: Davis King
Date: Sep 05, 2016 (05:54:31 UTC)

Renamed loss_binary_mmod_ to loss_mmod_

---

Revision: d263d2c5c848b166
Author: Davis King
Date: Sep 05, 2016 (05:37:30 UTC)

Added a version of test_object_detection_function() for the DNN based MMOD detector.

---

Revision: 5cdd56b819d8fc25
Author: Davis King
Date: Sep 05, 2016 (05:18:57 UTC)

Added an overload of upsample_image_dataset() for mmod_rect annotated datasets.

---

Revision: 4f0ff3268a1cfcdd
Author: Davis King
Date: Sep 05, 2016 (05:16:44 UTC)

Made input_layer() work in a more reasonable and general way.

---

Revision: e6e24f679a528b4f
Author: Davis King
Date: Sep 04, 2016 (18:01:12 UTC)

Made cropper not output boxes smaller than it's supposed to. Also cleaned up code a little.

---

Revision: b228a7c9b99f8e45
Author: Davis King
Date: Sep 04, 2016 (15:10:15 UTC)

Added random_cropper

---

Revision: cbe12b2a062b1eb1
Author: Davis King
Date: Sep 04, 2016 (13:41:27 UTC)

Added set_all_bn_running_stats_window_sizes() and also changed the default batch normalization running stats window size from 1000 to 100.

---

Revision: 6dbfea4bb22c2f9a
Author: Davis King
Date: Sep 04, 2016 (10:52:29 UTC)

Added an overload of load_image_dataset() that outputs directly to mmod_rect instead of rectangle.

---

Revision: 1654442af760f3ea
Author: Davis King
Date: Sep 04, 2016 (10:48:36 UTC)

Moved some code around, added documentation for the mmod loss layer and its attendant objects. Also fixed a minor bug in the loss layer.

---

Revision: afa3967afc6997a7
Author: Davis King
Date: Sep 04, 2016 (10:01:32 UTC)

Added missing assert

---

Revision: a4139ba9088a228e
Author: Davis King
Date: Sep 04, 2016 (08:01:49 UTC)

Changed imglab --resample so that it never changes the aspect ratio of an image.

---

Revision: a0d16e6dbf1439a3
Author: Davis King
Date: Sep 04, 2016 (07:55:30 UTC)

Fixed error in pyramid_rate() formula.

---

Revision: 2cd7f14c9947e0a6
Author: Davis King
Date: Sep 04, 2016 (05:41:38 UTC)

Added get_double_in_range() to dlib::rand.

---

Revision: 3107e008c2aae596
Author: Davis King
Date: Sep 03, 2016 (05:20:47 UTC)

Added MMOD loss layer

---

Revision: d37ff995dc506e08
Author: Davis King
Date: Sep 03, 2016 (05:02:53 UTC)

Added input_rgb_image_pyramid

---

Revision: d667089afd38fd21
Author: Davis King
Date: Sep 03, 2016 (04:39:58 UTC)

more clarification

---

Revision: 51a99ae538e25807
Author: Davis King
Date: Sep 03, 2016 (04:37:48 UTC)

clarified spec

---

Revision: f5fc678498dc52b9
Author: Davis King
Date: Sep 03, 2016 (04:05:30 UTC)

Made load_image_dataset() more generic. Now it can operate on arrays of images that use std::vector in addition to dlib::array.

---

Revision: 4600afc66b79c7e3
Author: Davis King
Date: Sep 03, 2016 (03:52:55 UTC)

Added input_tensor_to_output_tensor() and output_tensor_to_input_tensor() along with the mapping functions necessary at each layer to support these routines.

---

Revision: f3e40c3157b66be0
Author: Davis King
Date: Sep 03, 2016 (03:14:07 UTC)

Added visit_layers_backwards(), visit_layers_backwards_range(), and visit_layers_range().

---

Revision: d58fccff9e325fe8
Author: jpblackburn
Date: Aug 31, 2016 (18:13:39 UTC)

Add dnn_trainer::train_one_step iterator signature (#212) Add an overload of dnn_trainer::train_one_step that takes a pair of iterators rather than a std::vector.

---

Revision: 2f9c279aaa846ac9
Author: Davis King
Date: Aug 31, 2016 (07:34:04 UTC)

Changed pinv() so it interprets its tol argument relative to the largest singular value of the input matrix rather than as an absolute tolerance.

---

Revision: 443a68b4f7761e12
Author: Davis King
Date: Aug 30, 2016 (18:10:19 UTC)

workaround bugs in visual studio

---

Revision: 8717c62459e84fbb
Author: Davis King
Date: Aug 30, 2016 (17:46:26 UTC)

made the build script compile in newer versions of visual studio before older ones.

---

Revision: f7a509d6cb76d641
Author: Davis King
Date: Aug 30, 2016 (10:58:38 UTC)

Simplified example to show only the C++11 version of the code.

---

Revision: 19b311e139012a55
Author: Davis King
Date: Aug 30, 2016 (10:44:46 UTC)

Made thread_pool and parallel_for propagate exceptions from task threads to calling code.

---

Revision: 39a621bbba8eff85
Author: Davis King
Date: Aug 30, 2016 (06:15:33 UTC)

Added overloads of the parallel for functions that use default_thread_pool()

---

Revision: a6b6efd7482deb1d
Author: Davis King
Date: Aug 29, 2016 (19:34:02 UTC)

Added some comments

---

Revision: 60eff689078d877e
Author: Davis King
Date: Aug 29, 2016 (19:31:57 UTC)

Added a clone() method to the mex class interface.

---

Revision: 72e20e746c6fd3cc
Author: Davis King
Date: Aug 29, 2016 (10:15:24 UTC)

Fixed a problem where output matrices don't get assigned anything when they are empty, leading to MATLAB complaining about output arguments to being assigned.

---

Revision: 5fa3edc19eb93009
Author: Davis King
Date: Aug 29, 2016 (06:56:32 UTC)

Fixed tabbing

---

Revision: e8a4fa1094c2477f
Author: Davis King
Date: Aug 29, 2016 (05:46:27 UTC)

Added support for binding classes to MATLAB.

---

Revision: a05ea5995753e188
Author: Davis King
Date: Aug 28, 2016 (12:56:53 UTC)

Added pyramid_rate(), create_tiled_pyramid(), image_to_tiled_pyramid(), and tiled_pyramid_to_image().

---

Revision: 6de599a993c942f0
Author: Davis King
Date: Aug 28, 2016 (10:43:13 UTC)

Fixed compiler warning.

---

Revision: 27bad381c9e173d2
Author: Davis King
Date: Aug 28, 2016 (09:28:14 UTC)

Cleaned up assert statements a bit.

---

Revision: 5c74d9b3024f7989
Author: Davis King
Date: Aug 28, 2016 (08:09:02 UTC)

Added nearest_rect()

---

Revision: 2cc3419f93e3d4d3
Author: Davis King
Date: Aug 28, 2016 (08:07:27 UTC)

Made the message argument of the DLIB_ASSERT and DLIB_CASSERT macros optional.

---

Revision: 0ec79915109ddf79
Author: Davis King
Date: Aug 27, 2016 (16:13:19 UTC)

Moved impossible_labeling_error to error.h

---

Revision: d70d7afe43c38d10
Author: Davis King
Date: Aug 27, 2016 (12:38:34 UTC)

Added box_intersection_over_union() and also renamed the class members of test_box_overlap so they are less confusing and vague.

---

Revision: 607c51d65e523d4f
Author: Davis King
Date: Aug 27, 2016 (05:52:18 UTC)

Added alias_tensor_const_instance

---

Revision: c2cdd99160594c54
Author: Davis King
Date: Aug 27, 2016 (05:23:09 UTC)

Made add_loss_layer constructor more flexible. Now you can construct from objects that are implicitly convertible to a loss details object just like you can for computational layers.

---

Revision: 0d5e4cd99e388c3f
Author: Davis King
Date: Aug 27, 2016 (05:10:50 UTC)

Fixed a bug in memcpy() for tensors where you weren't allowed to copy alias tensors. Now any kind of tensors are supported.

---

Revision: 2b15e1983cd410ce
Author: Davis King
Date: Aug 24, 2016 (11:50:24 UTC)

Slightly adjusted some of the unit tests to avoid false failures in some environments.

---

Revision: d5c166fd952b9257
Author: Davis King
Date: Aug 24, 2016 (11:36:01 UTC)

Made test_layer() more numerically robust.

---

Revision: 5058bd86d75398fc
Author: Davis King
Date: Aug 24, 2016 (11:35:42 UTC)

Updated arch list to simd instructions are dealt with correctly on more platforms.

---

Revision: 17478a093d7f14c8
Author: Davis King
Date: Aug 24, 2016 (02:56:35 UTC)

fixed typo in example

---

Revision: 822f2dd0956a9230
Author: Davis King
Date: Aug 23, 2016 (13:01:36 UTC)

merged

---

Revision: 63e57910a46ec43c
Author: Davis King
Date: Aug 23, 2016 (12:52:24 UTC)

Minor changes to avoid compiler warnings

---

Revision: bda636fb12cfc45b
Author: Davis King
Date: Aug 23, 2016 (11:58:15 UTC)

Improved how the unit tests build all/source.cpp

---

Revision: 73c540c5f71a6ed9
Author: Davis King
Date: Aug 23, 2016 (11:05:55 UTC)

cmake cleanup

---

Revision: 548254133c54a418
Author: Davis King
Date: Aug 23, 2016 (02:41:19 UTC)

Adjusted test to avoid false alarm

---

Revision: 8bc9d6bcdff4299a
Author: Davis King
Date: Aug 22, 2016 (12:47:51 UTC)

Fixed rotate_image() so that it uses the generic image interface rather than expecting an array2d like object.

---

Revision: 24fbb5545ab5245f
Author: Davis King
Date: Aug 22, 2016 (11:08:20 UTC)

fixed compiler error

---

Revision: 8b7a8fd2688d8d1d
Author: Davis King
Date: Aug 22, 2016 (10:53:26 UTC)

Fixed typo in network definition.

---

Revision: 8e940b54b9872718
Author: Davis King
Date: Aug 22, 2016 (10:53:07 UTC)

Fixed compiler warnings

---

Revision: f85adb8c5a852e6c
Author: Davis King
Date: Aug 22, 2016 (07:36:56 UTC)

Made the thread local variables that hold the cudnn and cublas context objects not destruct and recreate themselves when you switch devices. Instead, they keep a table of context objects, for each thread and device, reusing as necessary. This prevents churn in the context objects when you are switching back and forth between devices inside a single thread.

---

Revision: 3c17706101b4654e
Author: Davis King
Date: Aug 22, 2016 (06:20:41 UTC)

Added cuda bindings: get_device_name(), set_current_device_blocking_sync().

---

Revision: 9980a9953beb5731
Author: Davis King
Date: Aug 22, 2016 (06:11:23 UTC)

Added is_row_major(). Also made sum() run over the matrix in column major order if that's the memory layout of the argument.

---

Revision: 116418659ea343b3
Author: Davis King
Date: Aug 22, 2016 (06:05:56 UTC)

Added an assert on the size of matlab's bools.

---

Revision: 2f0770cae6ca1c94
Author: Davis King
Date: Aug 21, 2016 (17:03:16 UTC)

Added annotation() to tensor so that you can associate any object you want with a tensor.

---

Revision: 6ee8d78f5d607b29
Author: Davis King
Date: Aug 21, 2016 (16:48:50 UTC)

Clarified spec

---

Revision: 6ccb35f11bacacb2
Author: Davis King
Date: Aug 21, 2016 (16:46:49 UTC)

Added a set_image_size() for sub_image_proxy objects so that they can be written to with assign_image(). However, you get an error if you try to assign an image with a different size.

---

Revision: 9dee7dbf0c7e13e4
Author: Davis King
Date: Aug 21, 2016 (15:57:18 UTC)

Added comments about thread safety requirements.

---

Revision: 0f97bad460adfff5
Author: Davis King
Date: Aug 21, 2016 (10:59:39 UTC)

Made the coordinate mapping between pyramid layers slightly more accurate.

---

Revision: 100c60cf14d3963a
Author: Davis King
Date: Aug 16, 2016 (14:19:26 UTC)

Fixed more compiler warnings

---

Revision: 42fe6ce6dc96f68a
Author: Davis King
Date: Aug 16, 2016 (14:15:59 UTC)

Fixed some compiler warnings

---

Revision: 64f7d8ac72900162
Author: Davis King
Date: Aug 16, 2016 (14:14:53 UTC)

Made the unit tests compile all/source.cpp.

---

Revision: 6385cfd74b2be869
Author: Davis King
Date: Aug 15, 2016 (17:02:42 UTC)

Removed concat1 and inception1 templates to avoid user confusion.

---

Revision: 69766a9232619917
Author: Davis King
Date: Aug 15, 2016 (14:27:06 UTC)

Made the relational operators constexpr so they don't accidentally cause compilation errors when they get pulled into the scope of template metaprogramming expressions.

---

Revision: 88086eb59bdb7fce
Author: Davis King
Date: Aug 15, 2016 (14:08:39 UTC)

updated makefile to give -std=c++11 flag

---

Revision: bd4cf46b3b1794bb
Author: Davis King
Date: Aug 03, 2016 (07:18:37 UTC)

added missing requirement.

---

Revision: 095890b2d0548ac5
Author: Davis King
Date: Aug 14, 2016 (15:07:30 UTC)

Changed the DNN API so that sample_expansion_factor is a runtime variable rather than a compile time constant. This also removes it from the input layer interface since the DNN core infers its value at runtime, meaning users that define their own input layers don't need to specify it anymore.

---

Revision: 5e58ca8313c143ec
Author: Davis King
Date: Aug 14, 2016 (09:48:18 UTC)

Made layer_details() part of the SUBNET interface so that user defined layer details objects can access each other. Also added the input_layer() global function for accessing the input layer specifically.

---

Revision: 7251a950690b418c
Author: Davis King
Date: Aug 14, 2016 (05:12:51 UTC)

Fixed dlib.range's iterator interface to work in python 3.

---

Revision: e8e2440b9a96ea72
Author: Davis King
Date: Aug 13, 2016 (10:09:22 UTC)

Record last changeset and set PATCH version to 99

---

Revision: ad6cd2a3bfd54d48
Author: Davis King
Date: Aug 13, 2016 (10:07:28 UTC)

Created release v19.1

---

Revision: 680e88eb8f297cee
Author: Davis King
Date: Aug 13, 2016 (08:57:21 UTC)

Added a bat file for running all the unit tests through different versions of visual studio.

---

Revision: 5cadadf542c69c4c
Author: Davis King
Date: Aug 13, 2016 (08:47:55 UTC)

New versions of boost and cmake need to be told to look in boost's library folder to find boost's libraries when building on windows. Bizarre.

---

Revision: 0ecf895b3704d7ee
Author: Davis King
Date: Aug 13, 2016 (05:03:16 UTC)

Fixed compiler error on ubuntu 16.04

---

Revision: d9047c3e80dbfe21
Author: Davis King
Date: Aug 13, 2016 (05:02:57 UTC)

Fixed grammar.

---

Revision: 8cdf5a12835e58b6
Author: Davis King
Date: Aug 11, 2016 (18:12:40 UTC)

Added workaround for a bug in cuDNN5.1 which causes cudnnGetConvolutionBackwardFilterAlgorithm() to pick invalid algorithms, resulting in cuDNN not working correctly.

---

Revision: 3b42397e5f6d0014
Author: Davis King
Date: Aug 10, 2016 (05:30:04 UTC)

Added additional search names for boost-python

---

Revision: 0df486296245ad1f
Author: Davis King
Date: Aug 09, 2016 (12:33:44 UTC)

Fixed a bug in the cuDNN binding that on rare occasions caused NaN outputs from batch normalization. The running mean and variance need to be initialized when calling cuDNN, even if the averaging factor is 1. I don't think this was the case prior to cuDNN v5, but it certainly is the case now. This patch fixes this.

---

Revision: 6a92b3903c5d22d2
Author: Davis King
Date: Aug 09, 2016 (06:48:48 UTC)

merged

---

Revision: 7a6e1198de1d01e9
Author: Davis King
Date: Aug 09, 2016 (06:36:38 UTC)

made unit test more repeatable

---

Revision: 9b4528cc0b9ff705
Author: Davis King
Date: May 25, 2016 (03:07:04 UTC)

made tests more repeatable

---

Revision: 42025ba5bd122b10
Author: Davis King
Date: Aug 06, 2016 (08:39:13 UTC)

Added more testing messages

---

Revision: 5c64d900cad38b6f
Author: Davis King
Date: Aug 06, 2016 (08:00:41 UTC)

CMake change to enable cuda use on ubuntu 16.06

---

Revision: a0640f0749627a3d
Author: Davis King
Date: Aug 05, 2016 (15:14:49 UTC)

added missing #include file

---

Revision: f9f02581bad17a74
Author: Davis King
Date: Aug 05, 2016 (04:13:45 UTC)

Fixed a few places where lab color space still referred to the channels as being signed rather than unsigned.

---

Revision: cbfeed2c5cc36bcb
Author: sutr90
Date: Aug 05, 2016 (12:59:07 UTC)

Added test for RGB to Lab to RGB conversion.

---

Revision: 48240945b04f98a5
Author: sutr90
Date: Aug 03, 2016 (16:48:59 UTC)

Added tests for Lab pixels.

---

Revision: baa9e138ca187117
Author: sutr90
Date: Jul 28, 2016 (16:07:23 UTC)

Fixed PNG saving.

---

Revision: 39a666b40da798c9
Author: sutr90
Date: Jul 27, 2016 (19:51:01 UTC)

Fixes Lab tests.

---

Revision: 30ca64a52f1cb5c2
Author: sutr90
Date: Jul 27, 2016 (19:47:42 UTC)

Fixed grayscale assignment.

---

Revision: 2e6d005d8ae72224
Author: sutr90
Date: Jul 27, 2016 (19:25:19 UTC)

Changed lab pixel range. Added clamping on conversion functions.

---

Revision: b8ddd208106126b5
Author: sutr90
Date: Jul 26, 2016 (24:11:56 UTC)

Added missing conversion between HSI and CIELab.

---

Revision: 5973e418ddfffaa4
Author: sutr90
Date: Jul 26, 2016 (23:45:52 UTC)

Added support for CIELab color space pixels.

---

Revision: 2f998a6c2b9fc7bc
Author: Davis King
Date: Aug 04, 2016 (17:12:00 UTC)

Added guards so dlib::async code isn't pulled in when using old versions of visual studio (since they don't support c++11).

---

Revision: dea3aecb792345f2
Author: Davis King
Date: Aug 04, 2016 (17:10:12 UTC)

Added check for initializer list support so code compiles in visual studio.

---

Revision: dec488b0a04b48ab
Author: Davis King
Date: Aug 04, 2016 (17:09:52 UTC)

Removed call to std::round() since visual studio 2012 doesn't support it.

---

Revision: ebab1ebf5895864b
Author: Davis King
Date: Jul 31, 2016 (05:49:21 UTC)

Fixed silly typo in script

---

Revision: 920eb067626995d9
Author: Davis King
Date: Jul 31, 2016 (05:46:10 UTC)

removed excessive error checking

---

Revision: 8592a327373922f7
Author: Davis King
Date: Jul 31, 2016 (05:11:52 UTC)

Made --resample use a 64bit hash of the image data in the file names to make it so there won't ever be any name conflicts in the cropped images.

---

Revision: 389426f54db08084
Author: Davis King
Date: Jul 30, 2016 (14:33:50 UTC)

Added some scripts that help with imglab XML file management.

---

Revision: 742e2bdfa301bb37
Author: Davis King
Date: Jul 30, 2016 (13:19:20 UTC)

Added --files

---

Revision: bb41846587dcc6ce
Author: Davis King
Date: Jul 30, 2016 (10:34:53 UTC)

Added --ignore option. Also improved how scale jittering happens in --resample, now the output images are always the same dimensions but the objects scale changes.

---

Revision: 45b26a2734add36a
Author: Davis King
Date: Jul 29, 2016 (05:03:05 UTC)

Made this not use stdin's file descriptor for data transfer between the processes since sometimes stdin was closed in some environments. Also cleaned up the code a little bit.

---

Revision: a0270d8c59e740fe
Author: Jan Rüegg
Date: Jul 26, 2016 (16:30:20 UTC)

Make filo compile with WIN32_LEAN_AND_MEAN defined (#169)

---

Revision: a221b991180b1218
Author: Davis King
Date: Jul 28, 2016 (15:09:07 UTC)

Changed code to avoid advancing iterator beyond end since some compilers complain about this (and it's technically not allowed in C++).

---

Revision: 81a2e542ba9ed3c0
Author: Davis King
Date: Jul 28, 2016 (13:59:31 UTC)

fixed grammar in comment

---

Revision: 804a7ed24b17f3ed
Author: Davis King
Date: Jul 23, 2016 (16:18:31 UTC)

Fixed potential division by zero.

---

Revision: 0d3e88de37146012
Author: Davis King
Date: Jul 23, 2016 (09:34:08 UTC)

fixed typo in comment

---

Revision: 23fc09e5d35fb92c
Author: Davis King
Date: Jul 22, 2016 (19:01:33 UTC)

Add guard so code compiles in old compilers.

---

Revision: 3fdb9ac329a28cd2
Author: Davis King
Date: Jul 22, 2016 (13:03:34 UTC)

Improved cmake warning messages about cuDNN

---

Revision: d2e246c66cf245c9
Author: Davis King
Date: Jul 22, 2016 (12:24:25 UTC)

Added dlib::async() and default_thread_pool()

---

Revision: 7cf1fd98846fee25
Author: Davis King
Date: Jul 22, 2016 (12:22:57 UTC)

Renamed something to avoid name clash with standard library.

---

Revision: 5a0b03db37fcd528
Author: Evgeniy Fominov
Date: Jul 22, 2016 (19:11:13 UTC)

Shape predictor trainer optimizations (#126) * Shape predictor trainer optimizations * Fixed performance leak in single thread mode & made VS2010 support

---

Revision: 8363dd6fefa23d82
Author: Davis King
Date: Jul 21, 2016 (18:32:48 UTC)

Made remove_unobtainable_rectangles() work with all scan_fhog_pyramid objects.

---

Revision: 4d582db4bb95cb8d
Author: Davis King
Date: Jul 21, 2016 (18:32:25 UTC)

Improved --extract-chips and --resample a bit.

---

Revision: cf183da2f015aad5
Author: Davis King
Date: Jul 21, 2016 (13:10:51 UTC)

Made it so the matrix only uses matlab's allocation routines for matrices that are directly the arguments of a mex function. This way, we avoid the problem of dlib::matrix objects being created in other threads that internally call into matlab and mess it up (since matlab is not thread safe in any sense).

---

Revision: c1b48f3aa17a53ea
Author: Davis King
Date: Jul 20, 2016 (11:33:45 UTC)

Fixed path error in this cmake file.

---

Revision: de575b215a3e16c2
Author: Davis King
Date: Jul 20, 2016 (16:28:08 UTC)

Added missing const.

---

Revision: 8a27cd5b0c42b1bd
Author: Davis King
Date: Jul 20, 2016 (11:12:32 UTC)

Made the env var CUDA_VISIBLE_DEVICES propagate to the child process

---

Revision: 0e022a150261af3a
Author: Davis King
Date: Jul 20, 2016 (06:19:46 UTC)

Minor change to avoid linker errors in some cases.

---

Revision: 5e9022be8efd95ac
Author: Minglangjun Li
Date: Jul 20, 2016 (30:05:30 UTC)

Fixes #128. Added support to discontiguous Numpy arrays (#155)

---

Revision: cb56f23ae90b1d7c
Author: Davis King
Date: Jul 16, 2016 (05:06:29 UTC)

Renamed iterator types to correctly reflect their requirements (i.e. not input iterators in the sense implied by the C++ standard but rather at least forward iterators).

---

Revision: a0468ff0f9378b70
Author: Davis King
Date: Jul 16, 2016 (04:58:17 UTC)

Improved error message

---

Revision: 729a452551a63e54
Author: vsooda
Date: Jul 12, 2016 (29:46:06 UTC)

fix shape_predictor dlib::searialize and dlib::desearialize error (#156) * fix shape_predictor dlib::searialize and dlib::desearialize error * declare shape_predictor searialize and desearialize inline * fix inline declare issue: declare inline before implement

---

Revision: 9fd60eec1b2de368
Author: Davis King
Date: Jul 10, 2016 (03:46:31 UTC)

Added checks to see if cblas.h has already been #included.

---

Revision: fc6e5577adf9270b
Author: Davis King
Date: Jul 07, 2016 (15:29:42 UTC)

Updated search paths for the Intel MKL so they will find newer versions of the MKL on windows.

---

Revision: 7763c50984e5209a
Author: Fm
Date: Jun 28, 2016 (15:47:49 UTC)

Merge branch 'master' of https://github.com/davisking/dlib

---

Revision: 794ff3ee4b92a489
Author: Fm
Date: Jun 21, 2016 (13:42:09 UTC)

DNN Visual Studio 2015 UP3 support

---

Revision: 95da024d50f9a1d9
Author: Davis King
Date: Jul 05, 2016 (03:03:38 UTC)

Made this cmake script always enable whatever partial C++11 support your compiler has.

---

Revision: 7e5d93b9241702b5
Author: Davis King
Date: Jul 04, 2016 (08:19:50 UTC)

Added --extract-chips option to imglab.

---

Revision: ad3195ff2ec9bd04
Author: Davis King
Date: Jul 04, 2016 (04:35:45 UTC)

Fixed bugs in the new --resample option

---

Revision: bd21b2c5fbdf39c5
Author: Davis King
Date: Jul 03, 2016 (15:26:49 UTC)

Added the --resample option

---

Revision: 6143ef322d8c2fa8
Author: Davis King
Date: Jul 03, 2016 (15:26:21 UTC)

Made it so that input<array2d> and input<matrix> layers can be converted between each other.

---

Revision: eb716b09234f9b0e
Author: Davis King
Date: Jul 03, 2016 (15:22:00 UTC)

Added rectangle_transform

---

Revision: 5a763c5bb65d6987
Author: Davis King
Date: Jun 28, 2016 (16:17:03 UTC)

Moved the auxiliary cmake into a common folder.

---

Revision: ed6f144a0a307167
Author: Davis King
Date: Jun 27, 2016 (06:17:48 UTC)

Fixed missing part of spec

---

Revision: b0f5471654baacc9
Author: Davis King
Date: Jun 25, 2016 (14:31:21 UTC)

Added some more comments

---

Revision: 45d625c505fbec4a
Author: Davis King
Date: Jun 25, 2016 (11:05:14 UTC)

Record last changeset and set PATCH version to 99

---

Revision: 7210589728f6d83f
Author: Davis King
Date: Jun 25, 2016 (11:03:30 UTC)

Created release v19.0

---

Revision: 657c608fec03e34a
Author: Davis King
Date: Jun 25, 2016 (10:16:43 UTC)

fixed tabbing

---

Revision: f54faf884ed106bf
Author: Davis King
Date: Jun 25, 2016 (10:00:38 UTC)

Changed URLs to point to dlib.net instead of sourceforge.net

---

Revision: 8e7a0b7f1b36fd89
Author: Davis King
Date: Jun 25, 2016 (09:03:12 UTC)

fixed compiler warning

---

Revision: 935447da3526dd72
Author: Davis King
Date: Jun 25, 2016 (08:31:59 UTC)

Changed code so the validation listing file doesn't have to be in the imagenet root folder.

---

Revision: d112ab11d2160ed3
Author: Davis King
Date: Jun 25, 2016 (08:28:00 UTC)

Fixed bug in apply_random_color_offset(), it computed slightly wrong outputs.

---

Revision: 349648ad122bf778
Author: Davis King
Date: Jun 25, 2016 (07:17:07 UTC)

fixed compiler warnings

---

Revision: 760b4cdb0a564200
Author: Davis King
Date: Jun 25, 2016 (05:47:36 UTC)

Minor formatting cleanup

---

Revision: 40636e2e9f9060d0
Author: Davis King
Date: Jun 25, 2016 (05:42:22 UTC)

fixed typo

---

Revision: bcfee9140836f78e
Author: Davis King
Date: Jun 25, 2016 (05:40:11 UTC)

Updated examples to refer to the correct file names.

---

Revision: 9f785654dc6b1746
Author: Davis King
Date: Jun 25, 2016 (05:34:53 UTC)

renamed examples

---

Revision: c497ce79d35ae74e
Author: Davis King
Date: Jun 25, 2016 (05:26:51 UTC)

Added the program that made the resnet model.

---

Revision: 534ebf664f47aefb
Author: Davis King
Date: Jun 24, 2016 (06:51:03 UTC)

Changed test to avoid failure in visual studio.

---

Revision: 52c3bd988a107cb2
Author: Davis King
Date: Jun 24, 2016 (-03:28:52 UTC)

Changed the message that cmake displays when opencv isn't found so users don't get confused.

---

Revision: 567abd857dc34a8d
Author: Davis King
Date: Jun 24, 2016 (-03:24:03 UTC)

Added check back in since visual studio doesn't really work with the C++11 code in dlib.

---

Revision: e8baacc72fc1b2be
Author: Davis King
Date: Jun 23, 2016 (05:42:38 UTC)

A minor change to avoid a compiler error in clang

---

Revision: c70732018c46b64e
Author: GurunathB
Date: Jun 23, 2016 (26:32:12 UTC)

Updated comments to specify the position of points (#141) Updated comments to specify the position of points on the face for 68 points generated by the face landmark detector.

---

Revision: c086d238f6b64dee
Author: Davis King
Date: Jun 22, 2016 (18:30:45 UTC)

Added some comments

---

Revision: b4e8042977086fdf
Author: Davis King
Date: Jun 22, 2016 (18:26:48 UTC)

Added an example showing how to classify imagenet images.

---

Revision: dc90a5abe21b758b
Author: Davis King
Date: Jun 22, 2016 (18:20:19 UTC)

Added input_rgb_image_sized.

---

Revision: 96d7741e06057500
Author: nxwhite-str
Date: Jun 22, 2016 (17:17:16 UTC)

Add detection threshold adjustment to object detection python interface (#140) * Add cmake option to use external libjpeg on Mac OS * Add adjust_threshold to python object detector * Add cmake option to use external libjpeg on Mac OS * Add adjust_threshold to python object detector * Revert "Add cmake option to use external libjpeg on Mac OS" This reverts commit 01f7fd13ea3f2b519312900333f9b68847fd1633. * Update detector example to set adjust_threshold

---

Revision: 2908d73bbbf5c26f
Author: Davis King
Date: Jun 22, 2016 (17:43:51 UTC)

Removed support for old serialization formats in many of the DNN objects. This is to clean up the code since it was getting somewhat complex and this is the last opportunity to do this kind of cleanup prior to the release of dlib v19.0. If you have saved network objects and want to convert them to the current format, then make sure you checkout the previous commit (labeled with tag before_dnn_serialization_cleanup) and then deserialize and serialize your network back to disk.

---

Revision: 4d6b102506bb9e2f
Author: Davis E. King
Date: Jun 22, 2016 (07:47:38 UTC)

Merge pull request #139 from e-fominov/dnn_vs2015_up3 DNN Visual Studio 2015 Update3 support

---

Revision: 14afea5ff3b87e1c
Author: Fm
Date: Jun 22, 2016 (21:09:26 UTC)

#pragma warning moved to dnn.h

---

Revision: ae8d58db4973c70b
Author: Fm
Date: Jun 22, 2016 (20:54:28 UTC)

removed wrong empty line

---

Revision: a633a6714320720b
Author: Fm
Date: Jun 22, 2016 (20:53:37 UTC)

removed comment form net printing

---

Revision: d404e2d676b8984a
Author: Fm
Date: Jun 22, 2016 (20:51:06 UTC)

#pragma warning for C4503 and /bigobj

---

Revision: 3c0ffd41063b88e4
Author: Fm
Date: Jun 22, 2016 (20:22:43 UTC)

Added compiler flags for VS compiling DNN samples without warnings

---

Revision: a40d55b25999dff6
Author: Fm
Date: Jun 22, 2016 (19:53:27 UTC)

Fixed layer constructors to support VS2015

---

Revision: 1251abd05a506bf0
Author: Fm
Date: Jun 21, 2016 (22:35:36 UTC)

removed commented code

---

Revision: 702f2732971ff655
Author: Fm
Date: Jun 21, 2016 (22:15:25 UTC)

Visual studio now compiles dnn_mnist_advanced, inception and dtest

---

Revision: 4da85b5391e84ac9
Author: Fm
Date: Jun 21, 2016 (13:49:01 UTC)

Visual Studio 2015 UP3 support

---

Revision: a9767ba602e13b1d
Author: Davis King
Date: Jun 22, 2016 (07:09:49 UTC)

Changed conv layer to use cross-correlation rather than convolution.

---

Revision: c380016096f46a88
Author: Davis King
Date: Jun 19, 2016 (04:55:42 UTC)

Made it so the image_display lets you add parts outside the object's bounding box.

---

Revision: 0aa0400268ffbade
Author: Davis King
Date: Jun 13, 2016 (08:25:44 UTC)

Made subprocess_stream echo stuff written to the child's cout to the parent's cout. Also just made it a little nicer with regard to error handling.

---

Revision: 5acc96d2156973bb
Author: Davis King
Date: Jun 11, 2016 (19:08:51 UTC)

Moved log1pexp() and randomize_parameters() from core.h into utilities.h

---

Revision: 10fd985ffb498a15
Author: Davis King
Date: Jun 11, 2016 (18:47:52 UTC)

Added net_to_xml()

---

Revision: 44d9fba2b8ffed0d
Author: Davis King
Date: Jun 11, 2016 (18:13:48 UTC)

Made layer() work on const networks.

---

Revision: 9bbee06740b5ca74
Author: Davis King
Date: Jun 11, 2016 (17:15:36 UTC)

Made mat() work on empty tensors.

---

Revision: 1d215f63e5f3cf9f
Author: Davis King
Date: Jun 11, 2016 (17:15:04 UTC)

Adjusted asserts to allow mat() to output empty matrices.

---

Revision: 8fae9e8b82cf0d1d
Author: Davis King
Date: Jun 11, 2016 (08:58:59 UTC)

Added visit_layers()

---

Revision: 6e3bc66aa6b93b34
Author: Davis King
Date: Jun 11, 2016 (07:56:04 UTC)

clarified spec

---

Revision: 05a5e5363690cbc4
Author: Davis King
Date: Jun 11, 2016 (07:54:44 UTC)

updated example comments

---

Revision: 54ddafdee10fdf93
Author: Davis King
Date: Jun 11, 2016 (07:54:32 UTC)

Added tag_id and made skip and add_prev layers print their id numbers when written to an output stream.

---

Revision: e148425be8ece75f
Author: Davis King
Date: Jun 11, 2016 (05:37:44 UTC)

cleanup

---

Revision: c5e3341d3f6a6adb
Author: Davis King
Date: Jun 11, 2016 (04:04:51 UTC)

Updated tests now that the resizable_tensor has an assignment mode that automatically sizes the tensor.

---

Revision: d04c251b841a4a2a
Author: Davis King
Date: Jun 11, 2016 (04:04:22 UTC)

Made the resizable_tensor's assignment operator work in a more sensible way.

---

Revision: 24dbc7dc8b677c6a
Author: Davis King
Date: Jun 11, 2016 (04:03:40 UTC)

Made fill_gaussian() default to a standard normal distribution.

---

Revision: 4c257702f516cbb9
Author: Davis King
Date: Jun 11, 2016 (03:39:56 UTC)

Made this bit of the code a little clearer.

---

Revision: 283bae2f1b502e95
Author: Davis King
Date: Jun 10, 2016 (17:26:12 UTC)

Removed this include statement since it causes errors on some systems.

---

Revision: e1c0dd50c8123cdf
Author: Davis King
Date: Jun 10, 2016 (16:52:10 UTC)

Made imglab assign a random color to boxes with labels.

---

Revision: d4da2bf6089d0438
Author: Davis King
Date: Jun 10, 2016 (16:49:20 UTC)

Made holding ctrl and clicking on an overlay rectangle set the color to the default value as well as the label to the default.

---

Revision: c5e398c5ad20a2f9
Author: Davis King
Date: Jun 10, 2016 (12:40:50 UTC)

Fixed minor bugs in join_rows() and join_cols(). They didn't work when one of the matrices was empty.

---

Revision: bfab04ee3fd00c15
Author: Davis King
Date: Jun 10, 2016 (08:42:02 UTC)

fixed compiler warning

---

Revision: ed3a22282755bfa9
Author: Davis King
Date: Jun 10, 2016 (07:00:53 UTC)

Made resizable_tensor construction and assignment from matrices automatically set the size of the tensor.

---

Revision: 7abc9cdc99ca5431
Author: Davis King
Date: Jun 07, 2016 (11:37:46 UTC)

fixed warning and typo in comment

---

Revision: dc6d459bdd37748f
Author: Davis King
Date: Jun 07, 2016 (08:08:40 UTC)

Fixed missing return statement

---

Revision: 95fcd08cfe9b8131
Author: Davis King
Date: Jun 07, 2016 (05:01:21 UTC)

Made CMake's search for cuDNN a little more broad

---

Revision: 769251e198953ca9
Author: Davis King
Date: Jun 07, 2016 (02:27:24 UTC)

Made tests less likely to false alarm.

---

Revision: a2c3f8205bd8bd6a
Author: Davis King
Date: Jun 06, 2016 (17:18:09 UTC)

Made this bit of code not look crazy

---

Revision: 51deedd8836012e1
Author: Davis King
Date: Jun 05, 2016 (12:39:10 UTC)

Fixed solvers so they apply the bias specific multipliers to the correct parts of the parameter vectors.

---

Revision: 0734470533650bc7
Author: AbdealiJK
Date: Jun 05, 2016 (28:68:09 UTC)

save_jpeg: Use TRUE instead of true In some verisons on jpeg, TRUE is an enum, and so `true` fails because it is not of the enum's type. Now, all the libjpeg calls use TRUE/FALSE. Fixes https://github.com/davisking/dlib/issues/129

---

Revision: 4df0b675c5585b34
Author: Davis King
Date: Jun 05, 2016 (03:45:15 UTC)

Made the steps without progress counter reset immediately upon changing the learning rate.

---

Revision: 45b0db5d2ba20909
Author: Davis King
Date: Jun 01, 2016 (02:51:46 UTC)

made tests more repeatable

---

Revision: 9017ee37672b6536
Author: Davis King
Date: May 31, 2016 (08:40:48 UTC)

fixed compile time error

---

Revision: 763f4c6984a81014
Author: Davis King
Date: May 31, 2016 (08:37:25 UTC)

Added subprocess_stream so that complex things can be isolated from MATLAB's shenanigans in a separate process.

---

Revision: f5f7f0af60574181
Author: Davis King
Date: May 31, 2016 (08:27:59 UTC)

Made the mex wrapper trap all std::exception derived exceptions rather than just dlib exceptions.

---

Revision: 64027e70595b7888
Author: Davis King
Date: May 31, 2016 (02:45:02 UTC)

Made imglab show the name of the current image in the title bar.

---

Revision: 38a6da59fda1fc3e
Author: Davis King
Date: May 30, 2016 (09:14:04 UTC)

minor cleanup

---

Revision: 22868c99e2671f1d
Author: Davis King
Date: May 30, 2016 (07:39:16 UTC)

clarified spec

---

Revision: dd8ebb421e62df84
Author: Davis King
Date: May 30, 2016 (05:24:19 UTC)

fixed typo

---

Revision: a37f304c93a471bf
Author: Davis King
Date: May 30, 2016 (05:17:46 UTC)

A little more cleanup in the spec

---

Revision: fcd9cac1209ad70b
Author: Davis King
Date: May 30, 2016 (04:54:02 UTC)

fixed typo

---

Revision: 9ff207d88a180dd7
Author: Davis King
Date: May 30, 2016 (04:50:49 UTC)

clarified spec

---

Revision: e2263ff9f026587b
Author: Davis King
Date: May 30, 2016 (04:50:28 UTC)

Clarified some parts of the example.

---

Revision: 98f6ba7640ae5482
Author: Fm
Date: May 29, 2016 (20:21:42 UTC)

Replaced sizeof... with variadic templates

---

Revision: 4a1e8f032a7503de
Author: Fm
Date: May 27, 2016 (17:49:11 UTC)

Removed friend declaration of dnn_tester from core.h

---

Revision: 20baf67694b60a0f
Author: Fm
Date: May 27, 2016 (12:55:15 UTC)

Changed concat syntax into concat1, concat2..., made dtest more readable::

---

Revision: e445f5915620bb31
Author: Fm
Date: May 26, 2016 (22:40:10 UTC)

Added layer access and printing examples to inception sample

---

Revision: 17828db2f61a2744
Author: Evgeniy Fominov
Date: May 26, 2016 (21:26:08 UTC)

Fixed dnn_tester in GPU mode for cpu_tensor test

---

Revision: 66435f70abd6303a
Author: Fm
Date: May 26, 2016 (20:51:44 UTC)

fixed cuda::copy_tensor

---

Revision: 63be501a69605ade
Author: Fm
Date: May 26, 2016 (20:43:54 UTC)

depth_group replaced with concat layer

---

Revision: ce9fd4d4cfe2fbe0
Author: Fm
Date: May 26, 2016 (20:15:56 UTC)

Merge branch 'master' of https://github.com/davisking/dlib into dnn_group_layer

---

Revision: ccae12930d226e99
Author: Fm
Date: May 17, 2016 (16:36:48 UTC)

depth layer: cuda concat/split moved to cpu/cuda files

---

Revision: 1b0d06e416ec8c11
Author: Fm
Date: May 17, 2016 (16:07:04 UTC)

Grouping layer added

---

Revision: 3710f9bf7d1b3224
Author: Davis King
Date: May 30, 2016 (03:35:25 UTC)

Cleaned up the contracts a little.

---

Revision: 01810065fb5a3a96
Author: Fm
Date: May 30, 2016 (12:25:23 UTC)

Added getter for trainer::train_one_step_calls

---

Revision: 014ef16a406d4985
Author: Davis King
Date: May 25, 2016 (15:12:36 UTC)

Made add_prev output a tensor with dimensions that are the max of each of the dimensions of its inputs rather than always outputting a tensor that has the dimensions of its immediate predecessors.

---

Revision: 5c6dc31e052bfcbc
Author: Davis King
Date: May 23, 2016 (18:01:47 UTC)

updated example

---

Revision: 026791e9404a4608
Author: Davis King
Date: May 23, 2016 (18:01:37 UTC)

Added bias learning rate and weight decay multipliers to bn_ layers

---

Revision: 02842d1ead9a1bd9
Author: Davis King
Date: May 23, 2016 (16:25:43 UTC)

Relaxed the requirements for calling find_min_box_constrained() and find_max_box_constrained(). Now the bounds can be empty for some variables.

---

Revision: 587ed2505faee4c2
Author: Davis King
Date: May 23, 2016 (15:57:53 UTC)

Changed code to avoid recreating thread_local cuda context objects.

---

Revision: 522937162ba182fb
Author: Davis King
Date: May 23, 2016 (02:54:55 UTC)

fixed broken tests

---

Revision: 4b0ba40bdee010ac
Author: Davis King
Date: May 22, 2016 (12:30:09 UTC)

Fixed a bug I introduced a minute ago.

---

Revision: 2d780c8ff054e5fa
Author: Davis King
Date: May 22, 2016 (12:14:10 UTC)

Fixed a bug in visit_layer_parameter_gradients() and visit_layer_parameters() caused by num_computational_layers being wrong when tax layers were placed as the first layer. These visit functions being wrong also caused multi-GPU support to not work on such networks.

---

Revision: a06033d239b68adf
Author: Davis King
Date: May 22, 2016 (11:49:40 UTC)

Changed the trainer threading code to use dlib::thread_pool instead of std::async() since std::async creates new threads with each invocation, which in turn causes objects with thread_local storage duration to be reconstructed each time. This is problematic because CUDA context objects for cublas and cudnn get reconstructed over and over, slowing things down and generally using more resources than should be used.

---

Revision: 635828e53bcfa8b5
Author: Davis King
Date: May 22, 2016 (09:17:10 UTC)

Cleaned up code a little and made the example use a better version of the architecture.

---

Revision: 516e7894b163d9af
Author: Davis King
Date: May 22, 2016 (06:30:15 UTC)

Fixing tests

---

Revision: 3d9334b28d23a9b2
Author: Davis King
Date: May 22, 2016 (06:26:23 UTC)

Made the batch normalization epsilon user settable rather than being hard coded.

---

Revision: 4e294c1002fb2c55
Author: Davis King
Date: May 22, 2016 (05:59:34 UTC)

Added learning rate and weight decay multipliers to the con_, fc_, and bn_ layers. Updated the solvers to support this.

---

Revision: bc894dae480c2f42
Author: Davis King
Date: May 21, 2016 (19:23:06 UTC)

Added more tests for the new affine_transform_range()

---

Revision: dc05bb90894d0679
Author: Davis King
Date: May 21, 2016 (19:22:26 UTC)

Added affine_transform_range() and another overload of affine_transform()

---

Revision: b208f4f16b9c260d
Author: Davis King
Date: May 21, 2016 (19:16:49 UTC)

Added get_learning_rate_multiplier() and get_weight_decay_multiplier() global functions.

---

Revision: 7b9b836fbf616591
Author: Davis King
Date: May 20, 2016 (04:29:39 UTC)

Added Johannes Huber's natvis file for visual studio.

---

Revision: 618761489d8a13f8
Author: Davis King
Date: May 18, 2016 (18:22:56 UTC)

Added an error message if a camera isn't available.

---

Revision: 4e5df0eb8c401a7d
Author: Davis King
Date: May 15, 2016 (15:56:55 UTC)

Made LIB_INSTALL_DIR only appear when building dlib as an installable library, not when using dlib in another cmake project.

---

Revision: c3279168c3c34c99
Author: Davis King
Date: May 15, 2016 (12:28:44 UTC)

Now when you print a network to cout it will include the output tensor sizes for each layer if you have passed a tensor through the next.

---

Revision: d784ee76714119be
Author: Davis King
Date: May 15, 2016 (11:07:04 UTC)

Renamed compute_loss() to compute_loss_value_and_gradient() in the loss interface.

---

Revision: 39a1fc83d8d55527
Author: Davis King
Date: May 15, 2016 (10:52:33 UTC)

Added comment to show how to deserialize a network.

---

Revision: 0e27a08f87456e17
Author: Davis King
Date: May 15, 2016 (10:48:06 UTC)

Added a function to dnn_trainer that lets you query the "steps without progress" estimate. I also renamed the get/set functions for the shrink amount to have a consistent name and use the word "factor" instead of "amount".

---

Revision: 7444fe767025d77d
Author: Davis King
Date: May 15, 2016 (10:36:02 UTC)

Added set_learning_rate_schedule() to dnn_trainer.

---

Revision: 39d21efc90ccb750
Author: Davis King
Date: May 15, 2016 (10:31:06 UTC)

clarified comments.

---

Revision: 871c2767b46c95e1
Author: Davis King
Date: May 14, 2016 (16:30:45 UTC)

Changed the solver interface to take the learning rate and the layer details object as an input. This allows the solvers to exhibit a more complex behavior that depends on the specific layer. It also removes the learning rate from the solver's parameter set and pushes it entirely into the core training code. This also removes the need for the separate "step size" which previously was multiplied with the output of the solvers. Most of the code is still the same, and in the core and trainer the step_size variables have just been renamed to learning_rate. The dnn_trainer's relevant member functions have also been renamed. The examples have been updated to reflect these API changes. I also cleaned up the resnet definition and added better downsampling.

---

Revision: 13b81b51041983ed
Author: Davis King
Date: May 13, 2016 (19:08:03 UTC)

Fixed the in-place layers so that they don't interfere with the operation of skip layers and add_prev style layers. In particular, now in-place layers only overwrite the gradient information in their child layer if they are operating in in-place mode. Otherwise, they add their gradients to their child layers. It should also be noted that it's safe for in-place layers to overwrite gradients when in in-place mode since their child layers are inaccessible when in-place layers operate in in-place mode. This prevents any other layers from trying to add to the child layer, thereby avoiding the potability of layer interference. So the bug this change fixes is that, when not in in-place mode the child layers are still accessible but in-place layers were *still* overwriting child gradients.

---

Revision: b8ca79904d4df660
Author: Davis King
Date: May 13, 2016 (05:01:24 UTC)

Made printing to cout or cerr check for matlab ctrl+c and throw an exception if so. This way, if you have a long running mex file it will be killable if it is periodically printing.

---

Revision: 60eaa594e8f9ed9a
Author: Davis King
Date: May 13, 2016 (04:50:25 UTC)

Added intel architecture detection so SSE/AVX isn't enabled when it shouldn't be.

---

Revision: 618d4d0caeea942f
Author: Davis King
Date: May 13, 2016 (04:35:22 UTC)

Made dropout layer initialize its random number generator from std::rand() rather than always using the same seed.

---

Revision: 0783d49a146c2dca
Author: Davis King
Date: May 13, 2016 (02:40:15 UTC)

Fixed a compiler error on some systems.

---

Revision: 40786b4dc5d40d45
Author: Davis King
Date: May 13, 2016 (02:32:18 UTC)

slightly relaxed tests to avoid false failure

---

Revision: cc68d1ade3dc40ef
Author: Davis King
Date: May 12, 2016 (15:40:24 UTC)

I accidentally turned the max pooling layers into average pooling a few days ago. This fixes that and turns them back into max pooling layers.

---

Revision: 6ff5d3ff3c55eea7
Author: Davis King
Date: May 12, 2016 (15:22:53 UTC)

Fixed network constructors so you can give an input layer for initialization without getting a compiler error.

---

Revision: 568c5b7a3c53ac37
Author: David Seifert
Date: May 10, 2016 (11:18:12 UTC)

* Add pkg-config file, allowing for build-system agnostic dependencies * Define LIB_INSTALL_DIR cache variable, allowing for multilib installations * Discover BLAS and LAPACK via pkg-config if possible * Fix incorrect toolchain variables in "dlib/test/makefile"

---

Revision: bb367dc2e9167075
Author: Davis King
Date: May 09, 2016 (17:46:04 UTC)

Fixed compiler warnings

---

Revision: c3961540917e8ead
Author: Davis King
Date: May 07, 2016 (10:30:42 UTC)

Made example use the "everything" version of avg pooling.

---

Revision: 7cd9a0f83797aaa2
Author: Davis King
Date: May 07, 2016 (10:30:18 UTC)

Added max_pool_everything and avg_pool_everything.

---

Revision: 372ff5b758a749b6
Author: Davis King
Date: May 07, 2016 (03:53:27 UTC)

Changed extract_fhog_features() back to returning matrix<double> since that's what it's contract says needs to happen (and what existing code expects).

---

Revision: f4220a92d0945ae6
Author: Davis King
Date: May 05, 2016 (02:58:52 UTC)

Made test error reporting more reliable

---

Revision: 50ac43e3dd2f8241
Author: Davis King
Date: May 04, 2016 (17:40:29 UTC)

Fixed avg pooling filter sizes to avoid errors with the new rules about non-one based strides.

---

Revision: eb8b88ef7fa28a51
Author: Davis King
Date: May 04, 2016 (17:39:31 UTC)

Improved error messages

---

Revision: e9e47cb7c080c0d1
Author: Davis King
Date: May 04, 2016 (17:32:59 UTC)

Pushed the padding parameters into the con_, max_pool_, and avg_pool_ interfaces. Also changed the default behavior when the stride isn't 1. Now the filters will be applied only to the "valid" part of the image.

---

Revision: 220cb636af22fc71
Author: Davis King
Date: May 04, 2016 (17:27:10 UTC)

Added missing ; to DLIB_TEST and DLIB_TEST_MSG macros now that they are required.

---

Revision: 729ca298c8af56ba
Author: Davis King
Date: May 04, 2016 (17:13:28 UTC)

fixed typo in comment

---

Revision: 90ea73cf501ccbf3
Author: Davis King
Date: May 03, 2016 (18:55:46 UTC)

Added a padding parameter to the pooling and convolution classes. Still need to expose it in the final layer interface.

---

Revision: 83e6c2039931488d
Author: Davis King
Date: May 03, 2016 (17:44:39 UTC)

Made test less likely to false alarm.

---

Revision: 4f7582a53f71129e
Author: Davis King
Date: May 03, 2016 (16:36:09 UTC)

Improved macros

---

Revision: fdbea5506e703c20
Author: Davis King
Date: May 02, 2016 (14:53:00 UTC)

fixed typo in spec

---

Revision: c3b4a21758840cb5
Author: Davis King
Date: May 01, 2016 (02:56:34 UTC)

Made dlib work with libgif v5

---

Revision: 954882c0e1988cf9
Author: Davis King
Date: Apr 30, 2016 (08:53:53 UTC)

Fixed bugs in multi-gpu training code.

---

Revision: 20cd0a18e869097d
Author: Davis King
Date: Apr 30, 2016 (08:28:00 UTC)

Added comments

---

Revision: 887d57118bd8fd5e
Author: Davis King
Date: Apr 30, 2016 (07:09:50 UTC)

Added a num_device_groups() member function.

---

Revision: 07892c7da7acd6cf
Author: Davis King
Date: Apr 30, 2016 (06:42:16 UTC)

Made launch_kernel() not generate CUDA errors if you ask it to launch a kernel of size 0.

---

Revision: 421efd4210679302
Author: Davis King
Date: Apr 29, 2016 (05:53:05 UTC)

Added missing #include.

---

Revision: 65cf7575aafb4cbd
Author: Davis King
Date: Apr 29, 2016 (05:40:19 UTC)

Fixed namespace lookup problem due to a name clash.

---

Revision: d62d1a0d6ffeb2fe
Author: Davis King
Date: Apr 29, 2016 (04:04:54 UTC)

Fixed compiler warnings

---

Revision: 571005f4faf0882f
Author: Davis King
Date: Apr 29, 2016 (02:55:53 UTC)

Made multi-gpu mode use GPUDirect rather than copying through the CPU.

---

Revision: 6fe15da64792f203
Author: Davis King
Date: Apr 29, 2016 (02:55:04 UTC)

Made the multi_device_tensor_averager not assume the size of the tensors is known at set() time.

---

Revision: 48a99971f1ea434c
Author: Davis King
Date: Apr 29, 2016 (02:54:30 UTC)

clarified spec

---

Revision: af950eda6373c303
Author: Davis King
Date: Apr 27, 2016 (17:50:31 UTC)

Added multi_device_tensor_averager

---

Revision: 382ca209c76c35ab
Author: Davis King
Date: Apr 28, 2016 (15:41:27 UTC)

clarified example

---

Revision: 4753e463b89f997b
Author: Davis King
Date: Apr 28, 2016 (14:57:15 UTC)

Fixed a bug in gaussian_blur() that caused messed up outputs when big sigma values were used on some pixel types.

---

Revision: 41c444435b2385a9
Author: Davis King
Date: Apr 27, 2016 (16:20:06 UTC)

Added more overloads of affine_transform()

---

Revision: c5214e7148b85015
Author: Davis King
Date: Apr 27, 2016 (16:18:57 UTC)

Added device_id() methods to gpu_data and tensor objects. These functions allow you to find out which device owns the memory inside these objects.

---

Revision: d977f0ca4678aeb6
Author: Davis King
Date: Apr 27, 2016 (08:36:50 UTC)

Made the cout and cerr redirection in mex files restore the state of the streambufs before exiting.

---

Revision: ea754bbf352d401d
Author: Davis King
Date: Apr 27, 2016 (04:49:02 UTC)

Fixed double counting of mini-batches for the purposes of solver termination when multiple GPUs are used.

---

Revision: d67d73abd4f07c93
Author: Davis King
Date: Apr 27, 2016 (04:48:23 UTC)

Fixed bad random number generation in layer initialization.

---

Revision: b71f11ff98266453
Author: Davis King
Date: Apr 26, 2016 (06:24:19 UTC)

Made the elastic_net inputs be in terms of trans(X)*X and trans(X)*Y rather than raw X and Y matrices.

---

Revision: 149bf8ce51ab635f
Author: Davis King
Date: Apr 26, 2016 (05:52:19 UTC)

made code faster

---

Revision: df16cfe84a9c8d84
Author: Davis King
Date: Apr 25, 2016 (13:05:10 UTC)

Added elastic_net solver.

---

Revision: 82b56e9009ce9f58
Author: Davis King
Date: Apr 25, 2016 (12:34:14 UTC)

removed cruft

---

Revision: c09b25e5f529b5e6
Author: Davis King
Date: Apr 19, 2016 (17:57:59 UTC)

Fixed file I/O error that happens only on windows.

---

Revision: 1289c4158a69a14e
Author: Davis King
Date: Apr 25, 2016 (04:26:29 UTC)

Added missing function

---

Revision: bc511d37a590fd3e
Author: Davis King
Date: Apr 25, 2016 (03:01:36 UTC)

fixed bug in tests

---

Revision: 96de7615a7de6fa9
Author: Davis King
Date: Apr 24, 2016 (06:08:55 UTC)

Added memcpy() functions for tensor and gpu_data objects.

---

Revision: dba3d6dfde580066
Author: Davis King
Date: Apr 24, 2016 (06:07:24 UTC)

Improved more macros

---

Revision: 36ec0f54919fda74
Author: Davis King
Date: Apr 24, 2016 (06:05:37 UTC)

Improved CHECK_CUDA() macro

---

Revision: 2d3266a41aa46fb6
Author: Davis King
Date: Apr 24, 2016 (04:33:39 UTC)

fixed grammar in comment

---

Revision: fb67a0f047dfce1e
Author: Davis King
Date: Apr 21, 2016 (13:27:56 UTC)

Fixed an error where some assignments to matrix output variables would result in an exception getting thrown.

---

Revision: a3cf389a0072b72d
Author: Davis King
Date: Apr 20, 2016 (04:39:49 UTC)

Made the mex wrapper flush output streams to the matlab console when the mex function ends.

---

Revision: 503feb0d491f3a18
Author: Davis King
Date: Apr 19, 2016 (02:45:35 UTC)

Added visit_layer_parameter_gradients() and also fixed a silly synchronization error in the multi-gpu training code.

---

Revision: 963c45cf80a9f8f9
Author: Davis King
Date: Apr 19, 2016 (02:44:31 UTC)

Fixed typo in example

---

Revision: 1fbe04b3b9053871
Author: Davis King
Date: Apr 18, 2016 (18:48:07 UTC)

Added comments about using multiple GPUs

---

Revision: cf6bbb04b4e881da
Author: Davis King
Date: Apr 18, 2016 (18:03:02 UTC)

Added multi-gpu support to the dnn_trainer

---

Revision: d764426ce0679f89
Author: Davis King
Date: Apr 18, 2016 (18:02:42 UTC)

Added missing assert

---

Revision: 1e568527c996bcaf
Author: Davis King
Date: Apr 18, 2016 (14:11:38 UTC)

Added find_optimal_parameters()

---

Revision: 1742063e07f8b432
Author: Davis King
Date: Apr 18, 2016 (14:11:16 UTC)

Fixed typo in error message

---

Revision: f6a281b9d2e007c8
Author: Davis King
Date: Apr 18, 2016 (14:04:00 UTC)

Added operator= that takes an initializer_list so that = assignments always behave in the way you would expect.

---

Revision: 15012d37a03bd5be
Author: Davis King
Date: Apr 18, 2016 (13:35:02 UTC)

Added a matrix constructor that takes an initializer list.

---

Revision: 729d47ae6f821e96
Author: Davis King
Date: Apr 17, 2016 (10:39:41 UTC)

Added visit_layer_parameters()

---

Revision: a3a8db1dbb4e78e2
Author: Davis King
Date: Apr 17, 2016 (10:35:40 UTC)

Clarified error message

---