Commit Graph

6109 Commits

Author SHA1 Message Date
Davis King 992dcd48a6 merged 2016-06-05 16:39:35 -04:00
Davis King aa19ff7377 Fixed solvers so they apply the bias specific multipliers to the correct parts
of the parameter vectors.
2016-06-05 16:39:10 -04:00
Davis E. King 02f6da2851 Merge pull request #130 from AbdealiJK/ajk/jpeg
save_jpeg: Use TRUE instead of true
2016-06-05 14:15:51 -04:00
AbdealiJK c102fb2690 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
2016-06-05 23:40:10 +05:30
Davis King 3c002d1cff Made the steps without progress counter reset immediately upon changing the
learning rate.
2016-06-05 07:45:15 -04:00
Davis King 72a2e8e437 made tests more repeatable 2016-06-01 06:51:46 -04:00
Davis King b230e7c33d fixed compile time error 2016-05-31 12:40:48 -04:00
Davis King ba59ddc6b5 Added subprocess_stream so that complex things can be isolated from MATLAB's
shenanigans in a separate process.
2016-05-31 12:37:25 -04:00
Davis King 0d2bce15ff Made the mex wrapper trap all std::exception derived exceptions rather than
just dlib exceptions.
2016-05-31 12:27:59 -04:00
Davis King 623fba97fe updated ignore list 2016-05-31 12:27:31 -04:00
Davis King 738b4d36af Made imglab show the name of the current image in the title bar. 2016-05-31 06:45:02 -04:00
Davis King 6e0f13ba06 minor cleanup 2016-05-30 13:14:04 -04:00
Davis King b4b9376aab updated docs 2016-05-30 13:04:23 -04:00
Davis King f698b85d68 clarified spec 2016-05-30 11:39:16 -04:00
Davis King f1eae955ac fixed typo 2016-05-30 09:24:19 -04:00
Davis King 20d10efc65 A little more cleanup in the spec 2016-05-30 09:17:46 -04:00
Davis King abd0019df0 fixed typo 2016-05-30 08:54:02 -04:00
Davis King 771ca2e0f3 clarified spec 2016-05-30 08:50:49 -04:00
Davis King 53e9c15811 Clarified some parts of the example. 2016-05-30 08:50:28 -04:00
Davis E. King 8c550d4c85 Merge pull request #114 from e-fominov/dnn_group_layer
Concat layer
2016-05-30 08:16:53 -04:00
Davis King cbd37d56a6 Cleaned up the contracts a little. 2016-05-30 07:35:25 -04:00
Davis E. King 7a31806baa Merge pull request #125 from e-fominov/dnn_trainer_get_step
Added getter for trainer::train_one_step_calls
2016-05-30 07:31:17 -04:00
Fm f06b265b34 Added getter for trainer::train_one_step_calls 2016-05-30 09:25:23 +03:00
Fm 01b3b08be6 Replaced sizeof... with variadic templates 2016-05-29 17:21:42 +03:00
Fm 1974e68d31 Removed friend declaration of dnn_tester from core.h 2016-05-27 14:49:11 +03:00
Fm d32bcdfa3d Changed concat syntax into concat1, concat2..., made dtest more readable:: 2016-05-27 09:56:00 +03:00
Fm 2f7d3578d2 Added layer access and printing examples to inception sample 2016-05-26 19:40:10 +03:00
Evgeniy Fominov 290b1cb15b Fixed dnn_tester in GPU mode for cpu_tensor test 2016-05-26 18:26:08 +03:00
Fm a06e533271 fixed cuda::copy_tensor 2016-05-26 17:51:44 +03:00
Fm 1f0318e222 depth_group replaced with concat layer 2016-05-26 17:43:54 +03:00
Fm 93e786db6c Merge branch 'master' of https://github.com/davisking/dlib into dnn_group_layer 2016-05-26 17:15:56 +03:00
Davis King 911638638d 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.
2016-05-25 19:12:36 -04:00
Davis King 66439bca25 made tests more repeatable 2016-05-25 07:07:04 -04:00
Davis King b9332698fe updated example 2016-05-23 22:01:47 -04:00
Davis King e5ad959085 Added bias learning rate and weight decay multipliers to bn_ layers 2016-05-23 22:01:37 -04:00
Davis King b6b8379819 Relaxed the requirements for calling find_min_box_constrained() and
find_max_box_constrained().  Now the bounds can be empty for some variables.
2016-05-23 20:25:43 -04:00
Davis King 974743767f Changed code to avoid recreating thread_local cuda context objects. 2016-05-23 19:57:53 -04:00
Davis King e55afabd1a fixed broken tests 2016-05-23 06:54:55 -04:00
Davis King 1cbf940eb3 Fixed a bug I introduced a minute ago. 2016-05-22 16:30:09 -04:00
Davis King f189612876 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.
2016-05-22 16:14:10 -04:00
Davis King d019e9cd08 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.
2016-05-22 15:49:40 -04:00
Davis King 5e70b7a2c6 Cleaned up code a little and made the example use a better version of the architecture. 2016-05-22 13:17:10 -04:00
Davis King b73dacc163 Fixing tests 2016-05-22 10:30:15 -04:00
Davis King 7f77ec6559 Made the batch normalization epsilon user settable rather than being hard coded. 2016-05-22 10:26:23 -04:00
Davis King b92b226c6a Added learning rate and weight decay multipliers to the con_, fc_, and bn_
layers.  Updated the solvers to support this.
2016-05-22 09:59:34 -04:00
Davis King 40f04bebae Added more tests for the new affine_transform_range() 2016-05-21 23:23:06 -04:00
Davis King c3a74c7c1c Added affine_transform_range() and another overload of affine_transform() 2016-05-21 23:22:26 -04:00
Davis King 15b2d7b5d8 Added get_learning_rate_multiplier() and get_weight_decay_multiplier() global
functions.
2016-05-21 23:16:49 -04:00
Davis King 58496f9f8a Added Johannes Huber's natvis file for visual studio. 2016-05-20 08:29:39 -04:00
Davis King 0cd76f899b Added an error message if a camera isn't available. 2016-05-18 22:22:56 -04:00