I’m sorry the example I shared before was flawed, I checked the source code the target size does expect the image size to be (img_height, img_width) and the Conv2d expects (img_height, img_width,img_channels)