Error training for x2

[davidjamesfriedman]

The size of tensor a (256) must match the size of tensor b (128) at non-singleton dimension 3

What do I need to change to fix this error?

[xinntao]

Here is the log of my training on X2 model. You can find the configs in the beginning.

train_002_MSRResNetx2_MSRResNetx4Ini_DIV2K_190609-174010.log

[davidjamesfriedman]

Thank you! I read other comments that say I need to change something in the discriminator to avoid this error as well. Which file should I modify to change this?

[xinntao]

Do you use the RRDB-Net ? If so, you need to comment two lines as said in #174 .

[davidjamesfriedman]

Yes I used RRDB-Net. I now get this error:

[davidjamesfriedman]

RuntimeError: Error(s) in loading state_dict for RRDBNet: untimeError: Error(s) in loading state_dict for RRDBNet: Unexpected key(s) in state_dict: "upconv2.weight", "upconv2.bias".

[xinntao]

If you load the pre-trained X4 model, try to set strict_load: false

[yl-to]

RuntimeError: Error(s) in loading state_dict for RRDBNet: untimeError: Error(s) in loading state_dict for RRDBNet: Unexpected key(s) in state_dict: "upconv2.weight", "upconv2.bias".

Did this solve your problem? If you did run it, could I know the steps please?Thanks!

[xinntao]

@Alongstory Did you want to train the X2 model? If so, this should work - set strict_load: false.

[yptheangel]

it worked for my case! @xinntao