OrgLion-ML
/
dlib
mirror of https://github.com/davisking/dlib.git
1
0
Fork 0
Code Issues Releases Wiki Activity
6,781 Commits 4 Branches 61 Tags 140 MiB
Commit Graph

6781 Commits

Author SHA1 Message Date
Davis King c3fa856c9f Fixed grammar 2017-08-22 22:02:19 -04:00
Davis King ebb0f85fac Fixed incorrect size() for simd8i. 2017-08-21 21:32:36 -04:00
Davis King d92728b334 Made windows testing bat file use rmdir rather than rm since rm isn't always available on windows. 2017-08-21 21:27:38 -04:00
Davis King f566b05a2e 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.
 2017-08-21 21:27:12 -04:00
Davis King 05c8391cb2 Fixed warning in visual studio. 2017-08-21 19:35:42 -04:00
Davis King 1db46949ee Made the test loss in the verbose output messages from the dnn_trainer not jump 
in variance when the learning rate resets.
 2017-08-20 20:47:00 -04:00
Davis King d009916e76 Added serialization support for the running_stats_decayed object. 2017-08-20 20:39:44 -04:00
Davis King dc45871a31 Made the loss value management a little more conservative. 2017-08-20 20:08:03 -04:00
Davis King dc071ceae1 Made the input_tensor_to_output_tensor() and output_tensor_to_input_tensor() 
coordinate mappings work on networks that contain skip layers.
 2017-08-20 19:42:12 -04:00
Davis King 620178db05 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.
 2017-08-20 19:30:11 -04:00
Davis King dd62b0e2ff 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.
 2017-08-20 19:28:08 -04:00
Davis King 618f1084d2 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.
 2017-08-20 07:14:40 -04:00
Davis King ba430be591 Make DLIB_ASSERT statements not abort the python interpreter, but just trigger an exception. 2017-08-19 08:48:48 -04:00
Davis King 33513abdeb Suppress compiler warning 2017-08-19 08:42:17 -04:00
Davis King 3156a5f440 merged 2017-08-18 21:24:49 -04:00
Adam Geitgey b6d2329c5e Add a python wrapper for using the mmod face detector (#753) 2017-08-18 16:30:33 -04:00
Davis King 2241e2c210 merged 2017-08-18 05:31:14 -04:00
Davis King 98cf44e8e3 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.
 2017-08-18 05:30:11 -04:00
Davis King 0863386849 merged 2017-08-15 15:52:16 -04:00
Davis King 48a56f36ab Fixed spelling error 2017-08-15 15:52:00 -04:00
Davis King 96619c859f merged 2017-08-15 09:32:26 -04:00
Davis King 4541a1b90a Fixed grammar 2017-08-14 21:36:18 -04:00
Davis King d6a1e273c0 Updated solvers to correctly pull in cont_'s bias parameter multipliers. 2017-08-14 14:04:16 -04:00
Davis King 9043c7415e Added extract_ layer 2017-08-14 13:50:47 -04:00
Davis King 525cfc71af Added more tests for copy_tensor() 2017-08-14 12:48:37 -04:00
Davis King 7078cfaff5 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.
 2017-08-14 12:28:26 -04:00
Davis King 89c9267e46 Made copy_tensor() use cudaMemcpyAsync() rather than cudaMemcpy(). 2017-08-14 09:52:53 -04:00
Davis King aafa411672 Added mult_prev layer. 2017-08-11 17:47:19 -04:00
Davis King f7310f4bbc Added multiply_zero_padded() 2017-08-11 16:39:00 -04:00
Davis King 46a02d9447 Made swig always run when you rebuild to avoid stale swig outputs. 2017-08-10 17:05:35 -04:00
Davis King 4ab360e439 Added DLIB_NO_ABORT_ON_2ND_FATAL_ERROR for dlib::fatal_error as a generic 
switch for use in plugin environments.
 2017-08-10 16:07:42 -04:00
Davis King 2630029cb4 Fixed missing java:: qualifiers. 2017-08-10 16:07:12 -04:00
Davis King e99d47cc7c Removed exit call on load library failure. 2017-08-10 15:31:02 -04:00
Davis King 44a62b19d0 A bit of path cleanup 2017-08-10 15:16:21 -04:00
Davis King 8f3249a438 merged 2017-08-10 14:38:59 -04:00
Davis King ae13ad800e Added more options for controlling the install folder paths. 2017-08-10 14:38:41 -04:00
Davis King af88b0d56f merged 2017-08-09 21:42:29 -04:00
Davis King d9f93548dd Fixed a bug in the warning message about NMS overlap where it would sometimes false alarm. 2017-08-09 21:42:14 -04:00
Davis King d62c5a7c8e merged 2017-08-09 12:20:06 -04:00
Davis King cf24f02507 Added an object that lets you hold a copyable reference to a java array. Also 
renamed the objects and generally improved documentation.
 2017-08-09 12:19:41 -04:00
Davis King fa5c666b6e Added an overload of mat() that takes a row stride value. 2017-08-08 15:10:17 -04:00
Davis King 594df9a66d Relaxes test to avoid false alarms. 2017-08-08 15:00:38 -04:00
Davis King 420eba0e6a Added note about logging training parameters. 2017-08-06 11:34:07 -04:00
Davis King 0f2753b754 Changed how we print the network hash. 2017-08-06 08:54:10 -04:00
Davis King 7b26b2d1ff Made dnn_trainer print the network size when logged to an iostream. 2017-08-05 15:24:47 -04:00
Davis King 8b21c89efd Improved how the relaxed mmod overlap settings are determined. 2017-08-04 22:46:46 -04:00
Davis King ed683785ce 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.
 2017-08-03 15:55:57 -04:00
Davis King 9540ca23ec Added operator<< for the DNN solvers. 2017-08-03 15:47:08 -04:00
Davis King aafde20607 Added operator<< for random_cropper. 2017-08-03 15:40:51 -04:00
Davis King a7d1dc474b Filled out the options for loss_mmod's operator<<. 2017-08-03 15:28:26 -04:00