KupynOrest/DeblurGAN

can not download weights

fatalfeel opened this issue · 7 comments

fatalfeel commented

Download weights from Google Drive . Note that during the inference you need to keep only Generator weights.
Put the weights into

chihina commented

I also have the same issue. Please anyone share a Generator model weight file!

fatalfeel commented

I will modify code let it simple and train it by myself, later paste here
DeblurGAN have v2 but its not original gan

Carrolinewang commented

I also have the same issue. Please anyone share a Generator model weight file!

Have you solve the problem? I have the same issue.

fatalfeel commented

i find out in DeblurGAN/checkpoints/experiment_name/
latest_net_D.pth
latest_net_G.pth
so you do not need to download, the latest_net_G.pth is the pretrain

Carrolinewang commented

i find out in DeblurGAN/checkpoints/experiment_name/
latest_net_D.pth
latest_net_G.pth
so you do not need to download, the latest_net_G.pth is the pretrain

I get it.Thank you.

fatalfeel commented

here i modify Perceptual-Loss that refer from

  1. https://github.com/tyui592/Perceptual_loss_for_real_time_style_transfer/blob/master/main.py
  2. https://github.com/rrmina/fast-neural-style-pytorch/blob/master/train.py
    change struct make DeblurGAN easy learn and train
    https://github.com/fatalfeel/DeblurGAN

later~ I will release FPN-Resnet to do G-net, its faster on cuda and do small block well

fatalfeel commented

we do not need to move to DeblurGANV2,
I add hi-resolution fpn152 with resnet-backbone here is the code.
https://github.com/fatalfeel/DeblurGAN/blob/master/models/fpn_resnet.py
Also add n_layers_G layers up options and n_blocks_G up options