pytorch-CycleGAN-and-pix2pix | https | Machine Learning library

torch.cat's first argument is expected to be a sequence of tensors rather than a single tensor. So you pass in like:

You can definitely have your model inherit from both the base class and torch.nn.Module (python allows multiple inheritance). However you should take care about the conflicts if both inherited class have functions with identical names (I can see at least one : their base provide the eval function and so to nn.module).

However since you do not need the CycleGan, and a lot of the code is compatibility with their training environment, you'd probably better just re-implement the pix2pix. Just steal the code, have it inherit from nn.Module, copy-paste useful/mandatory functions from the base class, and have everything translated into clean pytorch code. You already have the forward function (which is the only requirement for a pytorch module).

All the subnetworks they use (like the resnet blocks) seem to inherit from nn.Module already so there is nothing to change here (double check that though)

-unsafe won't help with this.

The problem is that your file names contain colons. Colons are not allowed in Windows file names.

Pytorch uses a tape based system for automatic differentiation. that means that it will backpropagate from the last operation it did. I think that the best way to understand is make a diagram from the process. I attach one that I did by hand

Now you will see that some modules are "repeated". The way I think about them is the same way I think about RNNs; in that way, the gradients will just be added.

Your observations are correct, but you may have missed the definition of UnetSkipConnectionBlock.forward() (UnetSkipConnectionBlock being the Module defining the U-Net block you shared), which may clarify this implementation:

(from pytorch-CycleGAN-and-pix2pix/models/networks.py#L259)

Some APIs out in the wild (such as inside setuptools) have similar kind of thing and they use it to their advantage. The __init__ call could be used for the low level internal API while public constructors are defined as classmethods for the different ways that one might construct objects. For instance, in pkg_resources.EntryPoint, the way to create instances of this class is to make use of the parse classmethod. A similar way can be followed if a custom initialization is desired