Output image size and cropping

[Walter5142]

hi @remicres thank you very much for this great work. I found that using the pre-training model to overscore the 128128 resolution image, the result is 256256 resolution. In addition, only a 64*64 region in the original image was intercepted for quadruple hyperpartitioning. Can you help me with the answer?

[remicres]

hi @Walter5142, Thanks. You are probably using the cropped output tensors. Take a look in train.py to see how thing are done !