nagadomi
382d493514
no confidence change for #128
2016-07-27 22:55:56 +09:00
nagadomi
b3c968ef11
Fix a bug in user method when scale=2
2016-07-20 14:06:17 +09:00
nagadomi
9563d84302
No bias at FullConvolution
2016-07-12 11:15:12 +09:00
nagadomi
81df729a8a
Remove -loss option
2016-07-09 15:05:11 +09:00
nagadomi
a0d8826779
Merge branch 'master' into upconv
2016-07-09 09:58:48 +09:00
nagadomi
8e90a44d30
noise0
2016-07-09 09:08:04 +09:00
nagadomi
06e073089b
Fix for edac608f
2016-07-06 08:04:24 +09:00
nagadomi
14330e919c
tuning a little
2016-07-06 08:03:24 +09:00
nagadomi
deae802267
Fix a bug in noise reduction when using Y model
2016-07-06 00:47:58 +09:00
nagadomi
edac608f18
Add support for user specified pairwise data for universal filter
2016-07-05 02:42:40 +09:00
nagadomi
9ec1f5159b
Fix a bug in ClippedMSECriterion
2016-07-04 11:16:04 +09:00
nagadomi
eeb90dc39b
Fix a bug in preprocessing when filters=box only
2016-06-22 00:37:29 +09:00
nagadomi
d8fcde46f3
Change default -validation_crops to 200.
2016-06-20 15:56:26 +09:00
nagadomi
fb3636c3b7
Change noise rate in jpeg training
2016-06-20 15:55:43 +09:00
nagadomi
f61d59d28a
Change default params. enabling jpeg_chroma_subsampling by default
2016-06-18 20:27:34 +09:00
nagadomi
9d3e1a241e
refactor
2016-06-18 20:25:15 +09:00
nagadomi
2e26122a9e
performance tuning
2016-06-17 20:40:03 +09:00
nagadomi
e50c7d1478
Remove unused modules
2016-06-13 12:48:11 +09:00
nagadomi
3b1b2d03fe
Change default -nr_rate to 0.65
2016-06-13 12:44:32 +09:00
nagadomi
25e293202a
Add support for tta_level=1; Add support for TTA to web.lua
2016-06-12 16:55:05 +09:00
nagadomi
af74a67bd1
Add -force_cudnn option; support for cuDNN in waifu2x.lua/web.lua
2016-06-12 16:33:50 +09:00
nagadomi
6be1479710
Fix a block_size issue when using upconv_7 model
2016-06-12 05:15:24 +09:00
nagadomi
9103d393fe
Fix a performance issue
2016-06-12 05:13:40 +09:00
nagadomi
b8ff8c6787
Remove -gamma_correction option
2016-06-10 07:37:39 +09:00
nagadomi
01b2e6d441
Remove -upsampling_filter option
2016-06-10 07:34:11 +09:00
nagadomi
afac4b52ab
Add -batch_size option to waifu2x.lua/web.lua
2016-06-09 14:03:18 +09:00
nagadomi
0b949c05a7
Add support for TTA level
2016-06-09 13:09:28 +09:00
nagadomi
9514027f65
Fix a bug that -nr_rate is not used
2016-06-09 02:44:22 +09:00
nagadomi
e5cfd3dfce
Add -resume option
2016-06-09 02:39:52 +09:00
nagadomi
51914b894a
change weight initialization and upconv_7
2016-06-08 06:58:46 +09:00
nagadomi
307ae40883
Add noise_scale training
2016-06-08 06:39:36 +09:00
nagadomi
d0630d3a20
individual filters and box-only support
2016-06-06 14:04:13 +09:00
nagadomi
0349fc774c
refactor
2016-06-02 10:12:04 +09:00
nagadomi
c89fd7249a
Add learning_rate_decay
2016-06-02 10:11:15 +09:00
nagadomi
70eb2b508f
Fix a performance problem in resampling
2016-05-30 19:15:54 +09:00
nagadomi
70a2849e39
Fix missing file
2016-05-30 06:48:26 +09:00
nagadomi
99e6dd1a57
Fix border removing
2016-05-28 10:25:08 +09:00
nagadomi
8a65db7bab
Change the evaluation metric
2016-05-27 16:57:14 +09:00
nagadomi
8088460a20
Add oracle_rate option
2016-05-27 16:54:29 +09:00
nagadomi
7814691cbf
Add resize_blur parameter
...
latest graphicsmagick is required
2016-05-21 09:54:12 +09:00
nagadomi
145b47dbf5
Add use_transparent_png option
2016-05-19 23:02:02 +09:00
nagadomi
f6a37b66c3
Add support for Transparent PNG in convert_data.lua
2016-05-15 12:34:03 +09:00
nagadomi
8d3950b90a
Change the default parameter (epoch, downsampling_filters)
2016-05-15 11:33:34 +09:00
nagadomi
c028ce6e4f
Fix a bug in reconstrct.scale() when inner_scale > 1 && y only model
2016-05-15 11:31:14 +09:00
nagadomi
a210090033
Convert model files; Add new pretrained model
...
- Add new pretrained model to ./models/upconv_7
- Move old models to ./models/vgg_7
- Use nn.LeakyReLU instead of w2nn.LeakyReLU
- Add useful attribute to .json
New JSON attribute:
The first layer has `model_config` attribute.
It contains:
model_arch: architecture name of model. see `lib/srcnn.lua`
scale_factor: if scale_factor > 1, model:forward() changes image resolution with scale_factor.
channels: input/output channels. if channels == 3, model is RGB model.
offset: pixel size that is to be removed from output.
for example:
(scale_factor=1, offset=7, input=100x100) => output=(100-7)x(100-7)
(scale_factor=2, offset=12, input=100x100) => output=(100*2-12)x(100*2-12)
And each layer has `class_name` attribute.
2016-05-15 03:04:08 +09:00
nagadomi
48411a4dde
refactor
2016-05-14 16:51:36 +09:00
nagadomi
51ae485cd1
Add new models
...
upconv_7 is 2.3x faster than previous model
2016-05-13 09:49:53 +09:00
nagadomi
e62305377f
Add compression.size()
2016-05-13 09:35:53 +09:00
nagadomi
958dcfaeb8
Add more informative message for CUDA error
2016-05-06 21:48:53 +09:00
nagadomi
5bfe504de1
Add error message when the CUDA module loading fails
2016-05-06 02:20:54 +09:00
nagadomi
8da52d5fb9
Merge from master
2016-04-23 12:48:24 +09:00
nagadomi
7af5c9443d
Add model option and 12 layers net
2016-04-23 09:19:03 +09:00
nagadomi
da03209d3e
Remove PNG compression option
2016-04-20 18:53:31 +09:00
nagadomi
68a6d4cef5
Use MSE instead of PSNR
...
PSNR depends on the minibatch size and those group.
2016-04-17 02:08:38 +09:00
nagadomi
3a27e122ac
Add support for grayscale output
2016-04-15 13:29:50 +09:00
nagadomi
ba2fe8d9c1
Improve alpha channel detection
...
latest `graphicsmagick` is required.
2016-04-15 13:14:10 +09:00
nagadomi
fbad30c031
Fix embed gamma handling
2016-04-15 09:13:37 +09:00
nagadomi
1407973b88
Use the specified filter
2016-04-12 01:33:17 +09:00
nagadomi
a938cd5994
Reduce draw calls
2016-04-10 23:30:23 +09:00
nagadomi
8d6451a51b
Add -max_training_image_size option
2016-04-10 23:06:39 +09:00
nagadomi
30fe5db735
Add upsampling_filter option
2016-04-02 22:03:27 +09:00
nagadomi
13f702b968
Add support for resizing with gamma correction
2016-03-28 19:07:09 +09:00
nagadomi
fd06a47d56
Fix clearState
2016-03-28 18:38:31 +09:00
nagadomi
7f448a98c4
Change default validation_crops(160)
2016-03-22 10:19:52 +09:00
nagadomi
cef98be88f
Change default downsampling_filters (Box,Catrom)
2016-03-22 10:18:59 +09:00
nagadomi
86ad50f7cd
Merge branch 'master' of github.com:nagadomi/waifu2x into dev
2016-03-21 03:59:17 +09:00
nagadomi
57e0f52b41
Reduce memory usage
2016-03-21 03:42:47 +09:00
nagadomi
8a799e2d56
Add support for noise level 3
2016-03-18 15:47:53 +09:00
nagadomi
4f67f47d3b
refactor
2016-03-17 20:03:04 +09:00
nagadomi
d4833160c7
Optionalize downsampling filters
2016-03-17 17:58:37 +09:00
nagadomi
4d115e4bdb
Add support for plotting loss chart
2016-03-14 05:06:14 +09:00
nagadomi
5590674783
Fix NaN in PSNRCriterion
2016-03-12 08:47:14 +09:00
nagadomi
4a1629d046
More clearState for nn.SpatialConvolutionMM
2016-03-12 07:25:16 +09:00
nagadomi
223dcead67
Add support for clearState() in w2nn.LeakyReLU
2016-03-12 06:59:23 +09:00
nagadomi
1900ac7500
Use PSNR for evaluation
2016-03-12 06:53:42 +09:00
nagadomi
41581a0d55
Remove cleanup_model
2016-03-12 05:43:04 +09:00
nagadomi
9b238bd693
Use clearState()
2016-03-11 11:12:02 +09:00
nagadomi
7708c7e75a
Merge branch 'master' of github.com:nagadomi/waifu2x
2015-12-11 16:35:00 +09:00
nagadomi
947746b32b
Change default parameter
2015-12-11 16:31:24 +09:00
nagadomi
56582b8401
Fix error in training script due to missing libcudnn.so
2015-12-09 07:04:04 +09:00
nagadomi
9f935835dd
Add -save_history option
2015-12-04 18:49:34 +09:00
nagadomi
eea4c31d7b
Change parameters
2015-12-04 18:47:33 +09:00
nagadomi
2305e31616
Merge branch 'master' into photo
2015-12-02 06:54:37 +09:00
nagadomi
03a17a636b
Merge branch 'master' of github.com:nagadomi/waifu2x
2015-12-02 06:54:22 +09:00
nagadomi
86feb1d4c9
Change the sampling method
2015-12-02 06:53:26 +09:00
nagadomi
9727bba51a
Fix gamma correction
2015-12-01 20:56:56 +09:00
nagadomi
5cda0b71d7
Fix weird value in alpha channel
2015-12-01 20:42:44 +09:00
nagadomi
c5df72439f
Fix cuda tensor error in async environment
2015-12-01 18:21:06 +09:00
nagadomi
d2c081bbcf
Improve alpha channel handling #29
...
- make border
- scale the alpha channel by waifu2x
- composite
2015-12-01 17:26:45 +09:00
nagadomi
a975155828
Change downscaling filters when photo training
2015-11-30 17:19:50 +09:00
nagadomi
aaac6ed6e5
Refactor training loop
...
more shuffle
2015-11-30 17:18:52 +09:00
nagadomi
c72ec3112b
Add -random_unsharp_mask_rate option for photo
2015-11-27 18:36:36 +09:00
nagadomi
d8ba661d6b
Add -jpeg_chroma_subsampling_rate option for JPEG denoise training
2015-11-26 17:10:57 +09:00
nagadomi
5a80a70e65
Fix error when image has alpha channel
2015-11-21 17:24:24 +09:00
nagadomi
abedec9d9b
Fix filter name for graphicsmagick
...
graphicsmagick package has bug in the definition of FilteTypes.
I will send PR to fix that bug.
This commit is related to it.
You should upgrade graphicsmagick package.
$ luarocks install graphicsmagick
2015-11-21 11:57:33 +09:00
nagadomi
f7b298690d
Fix the missing initialization function
...
I don't know why was this function removed.
2015-11-18 21:46:43 +09:00
nagadomi
b5db84d42e
Change the jpeg config for the photo model
2015-11-15 09:36:40 +09:00
nagadomi
42bd89151e
Add -gpu option in train.lua
2015-11-13 19:26:58 +09:00
nagadomi
c345f35ed0
Use Image.toString instead of Image.toBlob
2015-11-13 16:11:12 +09:00
nagadomi
3c1c11d88e
directly load float data
2015-11-13 15:53:41 +09:00
nagadomi
28cc864c0c
Fix NaN bug in iproc.white_noise
2015-11-12 08:03:24 +09:00
nagadomi
b63e08905b
Remove unused variable
2015-11-11 10:54:28 +09:00
nagadomi
4a4885c856
Add -white_noise option
...
post-processing for anime screencap.
2015-11-11 10:42:33 +09:00
nagadomi
0941d098e1
more round
2015-11-10 22:38:22 +09:00
nagadomi
2d5a090305
Change default -nr_rate
2015-11-10 22:07:45 +09:00
nagadomi
eaeaef10b7
more flip
2015-11-09 23:45:01 +09:00
nagadomi
af1b9c604b
refactor
2015-11-09 04:35:11 +09:00
nagadomi
b335f3a9ad
Add -tta option
...
The TTA mode:
- 8x slower than normal mode
- improves PSNR +0.1
2015-11-09 04:01:28 +09:00
nagadomi
7d67afb982
Use iproc.crop instead of image.crop
2015-11-08 22:11:36 +09:00
nagadomi
6e41796328
Use gamma 2.2 space when resizing
2015-11-08 20:28:14 +09:00
nagadomi
180e7676e4
Drop full transparent background
2015-11-08 06:32:15 +09:00
nagadomi
797b45ae23
Use roundf-like clip for 8 bit-depth image
...
Maybe PSNR +0.03 improved by this commit
2015-11-08 05:44:14 +09:00
nagadomi
bd63f99b59
clip
2015-11-08 01:21:52 +09:00
nagadomi
3b2aa67e55
Improve web interface
2015-11-07 22:54:29 +09:00
nagadomi
15f3f2d901
typo
2015-11-07 07:38:45 +09:00
nagadomi
84c38baab7
Fix double loading of image
2015-11-07 07:33:41 +09:00
nagadomi
3ea16b3b86
tunable parameters
2015-11-07 07:18:22 +09:00
nagadomi
cf862782a5
depth 8
2015-11-07 06:39:47 +09:00
nagadomi
c773e18e59
Add trade-off parameter for noise reduction
2015-11-07 06:37:53 +09:00
nagadomi
43832a9f92
Add support for depth of PNG
2015-11-06 13:12:36 +09:00
nagadomi
539941c234
Add support for url cache in web.lua
2015-11-06 12:24:27 +09:00
nagadomi
903d945652
cleanup
2015-11-06 10:08:54 +09:00
nagadomi
44208f18ab
Fix handling for gamma embed PNG
2015-11-05 14:59:51 +09:00
nagadomi
bbd0ff39ad
Fix handling for CMYK JPEG
2015-11-03 23:20:21 +09:00
nagadomi
b35a9ae7d7
tuning
2015-11-03 06:10:44 +09:00
nagadomi
490eb33a6b
Minimize the weighted huber loss instead of the weighted mean square error
...
Huber loss is less sensitive to outliers(i.e. noise) in data than the squared error loss.
2015-10-31 22:05:59 +09:00
nagadomi
243d8821be
Add -backend option to use cuDNN
2015-10-31 04:38:28 +09:00
nagadomi
9ddee6088d
Add WeightedHuberCriterion
2015-10-31 04:36:20 +09:00
nagadomi
9d63ed2947
Fix gradient scale in WeightedMSECriterion
2015-10-31 04:35:33 +09:00
nagadomi
4c691b4640
refactor
2015-10-29 22:44:15 +09:00
nagadomi
21ea5dd858
add cudnn converter
2015-10-29 18:05:33 +09:00
nagadomi
da786e15ba
remove noise_scale training
2015-10-28 16:27:31 +09:00
nagadomi
3abc5a03e3
refactor
2015-10-28 16:01:07 +09:00
nagadomi
956ae0d24c
merge from master
2015-10-26 09:39:41 +09:00
nagadomi
8dea362bed
sync from internal repo
...
- Memory compression by snappy (lua-csnappy)
- Use RGB-wise Weighted MSE(R*0.299, G*0.587, B*0.114) instead of MSE
- Aggressive cropping for edge region
and some change.
2015-10-26 09:23:52 +09:00
Kenta Murata
306ee3c76d
Fix typo
2015-08-30 10:47:20 +09:00
nagadomi
ee826715b7
explicitly loading torch/cutorch
2015-08-05 16:20:45 +09:00
nagadomi
2ffa93213e
remove exception handling for cunn
2015-08-05 16:03:51 +09:00
nagadomi
dcd43b89fd
fix missing nn in latest torch7
2015-08-05 11:49:38 +09:00
nagadomi
54580ba8c0
add data augmentation method that uses overlay
2015-08-02 22:02:14 +09:00
nagadomi
6c65463564
fix slow compression
2015-07-25 02:39:36 +09:00
nagadomi
e5a2ef97c7
set compression level of the output file
2015-07-24 17:16:27 +09:00
nagadomi
8e1625d001
change the noisy:cleaned ratio of training data
2015-07-18 11:17:34 +09:00
nagadomi
e3d3a8355c
change training script
...
- add AlexNet's color noise (default: false)
- add `photo` category for noise level setting
2015-07-11 21:57:04 +09:00
nagadomi
d8ca857a82
change initial weights
2015-07-11 14:52:51 +09:00
nagadomi
fd9dadd7a4
use settings.validation_crop
2015-06-26 20:12:51 +09:00
nagadomi
5b4d692f03
add support for RGB color space reconstruction
...
- add new RGB model (models/anime_style_art_rgb).
- RGB model can reduce color noise.
- waifu2x uses this RGB model by default.
You can use Y model with:
$ th waifu2x.lua -model_dir models/anime_style_art -i input.png -o output.png
$ th train.lua -color y ...
2015-06-23 02:55:30 +09:00
nagadomi
628bd971c9
Merge branch 'master' of github.com:nagadomi/waifu2x into rgb
2015-06-22 23:04:17 +09:00
nagadomi
04c1d7f936
remove unused function
2015-06-22 23:03:14 +09:00
nagadomi
1cdceb1648
throw error when failed to load image
2015-06-22 23:01:56 +09:00