search

ZJULearning / RMI

This is the code for the NeurIPS 2019 paper Region Mutual Information Loss for Semantic Segmentation.
MIT License
269 stars 37 forks source link

Effect of resizing image & mask without keeping width-height ratio in the data augmentation #40

Open rose-jinyang opened 2 years ago

rose-jinyang commented 2 years ago

Hello How are you? Thanks for contribution to this project. I'm NOT sure if this RMI loss would work well in case that we resize the image & mask to input size(NxN in pixels) without keeping width-height ratio in the data augmentation step. I am working on image segmentation project. There are many images & masks with different sizes in my dataset. The data by dataloader are resized to input size(ex: 256x256) and feed into the model. So the original width-height ratio of image & mask are NOT kept. Even in such case, does this RMI loss work well?

rose-jinyang commented 2 years ago

Hello Could u reply to the above question? Thanks

mzhaoshuai commented 2 years ago

Well, I think different w-h ratios will have no effect on RMI. RMI only cares about the relationship of neighbors, and resizing does not change it.

Sorry for this late reply. This repo is old and I am busy with other projects.

rose-jinyang commented 2 years ago

Thanks