speedinghzl / pytorch-segmentation-toolbox

PyTorch Implementations for DeeplabV3 and PSPNet
MIT License
768 stars 167 forks source link

question about maxpool #39

Open chenxiaoyu523 opened 5 years ago

chenxiaoyu523 commented 5 years ago

Thanks for your excellent work. I notice that the self.maxpool in your arch is redefined, where the ceiling mode is used. Can you explain why you use this mode and what result it would if the ceiling mode is false.

speedinghzl commented 5 years ago

This change inherits from Caffe-version deeplab which needs strict and specific input size. And it's unnecessary in Pytorch-version and it does not affect the performance. So I retain this change.