Semantic segmentation output “wave” like edges using PyTorch's DeepLabV3

Question

I\'m training a DeepLabV3 net from PyTorch and I was wondering if anyone can give me some tips regarding the "wave" shape of edges. Here\'s an output sample and I want