Awesome work! but can you provide the results of each class's mIoU of the original model and the improved model, which can help me think better about the problems that can be improved. And this work "Exploring Cross-Image Pixel Contrast for Semantic Segmentation"
Best.