RuntimeError: output with shape [1, 300, 200] doesn't match the broadcast shape [3, 300, 200] about instagan HOT 7 CLOSED

Hi, thank you for your suggestion. I modified requirements.txt to use == instead of >=.

from instagan.

Thanks for the super quick response! Great job on the paper btw. I'm using it for a class project.

from instagan.

Thank you for your suggestion.

I searched on the internet and finally got solved by downgrading torchvision from 0.4.1 to 0.2.1.

pip3 uninstall torchvision
pip3 install torchvision=0.2.1

Many of them suggest transforms.Normalize((0.5,), (0.5,)) but didn't help.
https://stackoverflow.com/questions/55124407/output-and-broadcast-shape-mismatch-in-mnist-torchvision
I'm not sure what was the original problem though..

from instagan.

Thanks for your interest! :)

from instagan.

Hi, it seems that the image type (RGB vs black/white) does not match in some reason.
Please see ./data/unaligned_seg_dataset.py to check if the loaded image is 1-channel and 3-channel, and also check the required channel for models.

from instagan.

I ran into the same issue. The torchvision version should be specified in requirements.txt and the conda script.

from instagan.

Hi, thank you for your suggestion.

from instagan.