birdortyedi / fashion-image-inpainting Goto Github PK

x_train = x_mask * y_train + (1.0 - x_mask) * 0.5

I've downloaded the pretrained weights but couldn't load it correctly. Could you spend some time making a demo script?

Hello, I'm very interested in this work. But I want to use this model on my own dataset. Would you mind teaching me how to train with a custom dataset? What kind of labels do I need to train this model? And are there any pre-trained models I can download?

Tried to input x_mask as a single channel binary mask with 1's at the area to be filled in with size torch.Size([1, 1, 256, 256])
however got

RuntimeError: Given groups=1, weight of size [64, 3, 7, 7], expected input[1, 1, 256, 256] to have 3 channels, but got 1 channels instead

What exactly is x_mask?

Excellent project! Do you have a pre-trained model available to test?

Tried to load the weights_net_df2.pth weights given. Firstly had to rename dict items to remove 'module' from parameter name stem but even then so getting Unexpected keys error'.
Here is how the loading was done and error:

from collections import OrderedDict
w=torch.load('weights_net_df2.pth')
new=OrderedDict({})

for key in w.keys():
  new_name=('.').join(key.split('.')[1:])
  new[new_name]=w[key]

net=Net()
net.load_state_dict(new)

Gives:

How to load weights correctly?

What is the function of 0.5 here? Is the area to be filled in given by 0.5 values?

x_train = x_mask * y_train + (1.0 - x_mask) * 0.5

Results in ReadMe look great. Can you share your pre-trained model for us to try as well?