Some questions

[Bellahanxi01]

What a great work！Your work has been a great inspiration to me. I encountered three issues during my experiments:

1.Pretrained weights: I couldn’t find the segformerb4_512x512ade160k.pth file, but I found segformerb4.pth. Can I rename it to segformerb4_512x512ade160k.pth and use it in your code?
2.If this is possible, does the image size need to be set to 512x512?
3.I’m doing binary classification, and I noticed that the output prediction image is completely black. What could be the reason? My images are in PNG format with 3 channels, and the masks are PNG format with a single channel and 8-bit values (only 0 and 255). I’ve set num_classes = 2 to match your code, and I’ve named my dataset "polyp" to directly use your code.

Thank you very much for your response!

[Bellahanxi01]

Sorry to bother you,I've found that the output from the model is all negative for pred = predictions[0, 1, :, :], is that normal?