4fad635aa4
Fix testing
2016-07-13 19:28:11 +02:00
2582f43797
dataset.lua: Comment out unused torch.linspace.
2016-07-12 12:06:12 -04:00
6534788a41
Use torch.range instead of torch.linspace because of rounding issues.
...
For #160
2016-07-12 11:45:15 -04:00
4c0e55dfd5
Rebase TripletEmbedding to get speedup at forward pass
2016-07-12 08:02:53 +02:00
0995e86b7e
Added Multi-GPU support #106
2016-07-04 22:23:37 +02:00
e489b7a7b8
Training: Correct space after device param when testing.
2016-06-19 23:42:15 -04:00
b82f6c31d6
Training: Pass device option to batch-represent.
2016-06-19 21:44:00 -04:00
0574a72641
Initial resnet model definition.
2016-06-19 17:13:21 -04:00
2c6ddfba2d
Training: Model: Use opt.imgDim.
2016-06-16 20:12:21 -04:00
1fb8abc247
Training: Improve -data description.
2016-06-14 15:56:37 -04:00
92684ab771
Training: Add cuda device option.
2016-06-14 15:55:46 -04:00
a1b3251f96
Actually fix nans from #127 . Error if they appear.
2016-06-14 13:51:18 -04:00
c04386c48c
Training: Only convert to cuda in cuda mode.
2016-06-13 14:12:36 -04:00
f4567e882e
Fixes for travis.
2016-06-13 14:06:12 -04:00
fea2fb41d8
Resolve nan issues for #127 (hopefully)
2016-06-13 13:50:38 -04:00
aaab8fe119
Training: Update plot-loss to new directory structure.
2016-06-13 13:50:38 -04:00
c6f0962b6d
Training: Add -save option. Default to a timestamp instead of numbers.
2016-06-13 13:50:38 -04:00
bced2bb445
Disable cudnn by default for #127 . Add warning.
2016-06-07 18:20:46 -04:00
21f1bcc5c4
Temporarily remove optimize_net for #127 .
2016-06-07 18:14:13 -04:00
130edcecc9
Training: Use ADAM instead of adadelta.
2016-06-03 17:48:38 -04:00
b4df1f9644
Revert dataset.lua for #132 .
2016-05-08 12:11:59 -04:00
0707db5718
Fix maxPathLength assertion/increment.
...
Thanks for the correction @aaronnech, I read your earlier suggestion too fast :-)
For #132
2016-05-07 20:43:53 -04:00
85d7375fd9
Remove unused variables/functions to fix luacheck warnings.
...
\cc @aaronnech
For #132
2016-05-07 20:33:52 -04:00
b11910b1da
Fix assert bound.
...
\cc @aaronnech
For #132
2016-05-07 20:30:17 -04:00
1d5491a1e8
Add @aaronnech's dataset.lua improvements for #132 .
2016-05-07 20:25:27 -04:00
eba58604c9
Bump matplotlib dependency for 'style'.
...
https://groups.google.com/forum/#openface/vJFioAQaJo8
2016-04-06 10:44:51 -04:00
66efcb706b
Model: Use opt.imgDim for retraining
...
\cc @melgor
Ideally I should have made imgDim part of the model.
This works for now, but users training will need to be
careful about this size.
2016-04-05 18:02:45 -04:00
1b3b48f588
Load util before model for optimizeNet
...
\cc @melgor
2016-04-03 02:48:50 -04:00
505c63f828
Added check if 'optnet' package is installed
2016-04-01 13:04:25 +02:00
46798c97cc
Change variable to local
2016-03-31 12:00:17 +02:00
bf6a466484
Added optnet for reducing memory consumption
2016-03-30 23:21:13 +02:00
6c43f6ba38
Training: Pass imgDim to batchRepresent. Add lfwDir option.
...
For #112 .
2016-03-19 15:44:47 -04:00
b30d6d9461
Training Plotting code: Add minor gridlines.
2016-03-18 14:10:53 -04:00
bba911c802
Replace satitize by clearState()
2016-03-16 18:20:11 +01:00
e7c574829d
Merge pull request #108 from melgor/cudnn_convert
...
Replace nn_to_cudnn by cudnn.convert #106
2016-03-16 11:54:20 -04:00
7d21caf004
Merge branch 'cudnn_convert' of https://github.com/melgor/openface into cudnn_convert
...
Conflicts:
training/model.lua
training/train.lua
2016-03-16 16:15:05 +01:00
f63acb637e
Replace nn_to_cudnn by cudnn.convert #106
2016-03-16 15:38:06 +01:00
6fd4ac8604
Training: Sample from all of a person's images.
...
This fixes a bug in the training code that caused
only the first (same) `nSamplesPerClass` images in each
class to be sampled every time rather than sampling
from all of the person's images.
2016-03-15 16:27:09 -04:00
e2d7513ff5
Replace nn_to_cudnn by cudnn.convert #106
2016-03-11 11:09:23 +01:00
71b6bbf765
Fix flake warning.
2016-03-08 20:39:03 -05:00
665cb4c9d7
Training plot-loss.py: Improve triplet loss plot bounds.
2016-03-08 18:36:58 -05:00
04d23481fb
Fix typo.
2016-03-06 21:15:10 -05:00
4616f61ff8
DNN Training: Halve default epochSize.
2016-03-06 20:03:54 -05:00
f17a36c2fd
plot-loss: Import sys.
2016-03-06 19:56:12 -05:00
f112398178
DNN Training: Plot LFW accuracies for #100 .
2016-03-06 19:46:33 -05:00
bbff0a01dd
Fix tests for training the DNN.
2016-03-06 19:45:37 -05:00
f5766412d2
NN Training: Add LFW validation for #100 .
2016-03-04 18:30:39 -05:00
e4c4aef3e4
Training: Zero pad experiment work directory numbers.
2016-03-04 18:09:55 -05:00
540b7d03c6
Fix luacheck warning.
2016-02-28 15:08:50 -05:00
49d195355b
Training: Remove test code from data loader.
2016-02-28 13:48:22 -05:00
melgor