I tried to train a pix2pix model on the edges2shoes-r dataset using train_full.sh.

Could you provide some more information? What is the type of your <code class="notrans

I try to print m1 and m2 type, the result look like this. <div class="snippet-clip

Distill Problem about gan-compression HOT 4 CLOSED

saijo0404 commented on June 11, 2024

Distill Problem

from gan-compression.

Comments (4)

lmxyy commented on June 11, 2024

Could you provide some more information? What is the type of your m1 and m2?

from gan-compression.

saijo0404 commented on June 11, 2024

I try to print m1 and m2 type, the result look like this.

distiller [ResnetDistiller] was created
Load network at logs/pix2pix/edges2shoes-r/train/checkpoints/latest_net_G.pth
isinstance(netA, nn.DataParallel):  False
isinstance(netB, nn.DataParallel):  False
isinstance(m1, nn.Conv2d) and isinstance(m2, (nn.Conv2d, SuperConv2d)):  True
m1 type:  <class 'torch.nn.modules.conv.Conv2d'>
m2 type:  <class 'torch.nn.modules.conv.Conv2d'>
isinstance(m1, nn.Conv2d) and isinstance(m2, (nn.Conv2d, SuperConv2d)):  True
m1 type:  <class 'torch.nn.modules.conv.Conv2d'>
m2 type:  <class 'torch.nn.modules.conv.Conv2d'>
isinstance(m1, nn.Conv2d) and isinstance(m2, (nn.Conv2d, SuperConv2d)):  True
m1 type:  <class 'torch.nn.modules.conv.Conv2d'>
m2 type:  <class 'torch.nn.modules.conv.Conv2d'>
isinstance(m1, nn.Conv2d) and isinstance(m2, (nn.Conv2d, SuperConv2d)):  False
m1 type:  <class 'models.modules.resnet_architecture.resnet_generator.ResnetBlock'>
m2 type:  <class 'models.modules.resnet_architecture.resnet_generator.ResnetBlock'>

from gan-compression.

lmxyy commented on June 11, 2024

I see. This is a minor bug in weight_transfer.py because of a typo. I've fixed it in this commit. Could you pull the latest commit and try again?

from gan-compression.

lmxyy commented on June 11, 2024

I will close this issue. Let me know if there are some further issues!

from gan-compression.

Recommend Projects