Memory-Based Cross-Image Contexts for Weakly Supervised Semantic Segmentation

This is the implementation of the method described in paper: Memory-Based Cross-Image Contexts for Weakly Supervised Semantic Segmentation.

Requirements

• Pyhton3.7+
• Pytorch1.0+
• Numpy, OpenCV
• Pydensecrf

Usage

The script main.py is self-contained, which calls scripts in other train_eval_xxx.py files for training. By defualt, simply run:

python main.py --gpus 0,1,2,3