 nagadomi
|
a210090033
Convert model files; Add new pretrained model
|
9 lat temu |
|
nagadomi
|
51ae485cd1
Add new models
|
9 lat temu |
|
nagadomi
|
b8088ca209
Remove the limit of learning_rate_decay
|
9 lat temu |
|
nagadomi
|
7af5c9443d
Add model option and 12 layers net
|
9 lat temu |
|
nagadomi
|
68a6d4cef5
Use MSE instead of PSNR
|
9 lat temu |
|
nagadomi
|
fa9355be7c
Fix validation metric
|
9 lat temu |
|
nagadomi
|
a92b2cb386
Fix progressbar
|
9 lat temu |
|
nagadomi
|
ea780f1871
Fix validation metric
|
9 lat temu |
|
nagadomi
|
30fe5db735
Add upsampling_filter option
|
9 lat temu |
|
nagadomi
|
13f702b968
Add support for resizing with gamma correction
|
9 lat temu |
|
nagadomi
|
d4833160c7
Optionalize downsampling filters
|
9 lat temu |
|
nagadomi
|
4d115e4bdb
Add support for plotting loss chart
|
9 lat temu |
|
nagadomi
|
459c7c5e18
Use similar distribution on train/validation
|
9 lat temu |
|
nagadomi
|
1900ac7500
Use PSNR for evaluation
|
9 lat temu |
|
nagadomi
|
664322fa97
Remove cleanup_model.lua; Change model format from binary to ascii
|
9 lat temu |
|
nagadomi
|
290a5f960b
Use Huber loss instead of MSE
|
9 lat temu |
|
nagadomi
|
9b238bd693
Use clearState()
|
9 lat temu |
|
nagadomi
|
9f935835dd
Add -save_history option
|
10 lat temu |
|
nagadomi
|
aaac6ed6e5
Refactor training loop
|
10 lat temu |
|
nagadomi
|
c72ec3112b
Add -random_unsharp_mask_rate option for photo
|
10 lat temu |
|
nagadomi
|
d8ba661d6b
Add -jpeg_chroma_subsampling_rate option for JPEG denoise training
|
10 lat temu |
|
nagadomi
|
42bd89151e
Add -gpu option in train.lua
|
10 lat temu |
|
nagadomi
|
5ab2c605c8
Fix header include order issues
|
10 lat temu |
|
nagadomi
|
797b45ae23
Use roundf-like clip for 8 bit-depth image
|
10 lat temu |
|
nagadomi
|
3ea16b3b86
tunable parameters
|
10 lat temu |
|
nagadomi
|
c773e18e59
Add trade-off parameter for noise reduction
|
10 lat temu |
|
nagadomi
|
903d945652
cleanup
|
10 lat temu |
|
nagadomi
|
b35a9ae7d7
tuning
|
10 lat temu |
|
nagadomi
|
490eb33a6b
Minimize the weighted huber loss instead of the weighted mean square error
|
10 lat temu |
|
nagadomi
|
da786e15ba
remove noise_scale training
|
10 lat temu |
