search

qibao77 / CFSNet

pytorch code of "CFSNet: Toward a Controllable Feature Space for Image Restoration"（ICCV2019）
62 stars 4 forks source link

denoising train #6

Open dorbarber79 opened 4 years ago

dorbarber79 commented 4 years ago

hello and thank you for your nice work. during training of gray denoise the model is not converge if i use α_in= 0. only if i use α_in= 1 the loss converges and reduce I tied several datasets and also with per-trained model and without.

thanks in advance for your help

dorbarber79 commented 4 years ago

in your denoise instructions you write: step 1 noise_level 25 no pre-trained model input_alpha 0

step 2 noise_level 50 use pre-trained model input_alpha 1

my problem is that in step 1 (input_alpha 0 ) the model loss is not reducing

qibao77 commented 4 years ago

I‘m sorry for late reply, has your problem been solved? I ran the code again in the denoising task, no problem. Can you provide more details?