Reproduce “Exploring Pixel-level Self-supervision for Weakly Supervised Semantic Segmentation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sung-Hoon Yoon, Hyeokjun Kweon, Jaeseok Jeong, Hyeonseong Kim, Shinjeong Kim, Kuk-Jin Yoon
Researchers at KAIST developed a framework for weakly supervised semantic segmentation that utilizes pixel-level self-supervision to overcome the sparsity and impreciseness of Class Activation Maps (CAMs). Their method achieves state-of-the-art performance on PASCAL VOC 2012, reaching 70.9% mIoU on the test set using only image-level labels.
Key insights
- Pixel-level self-supervision, derived internally from weak image-level labels, effectively refines CAMs without needing external datasets or pre-trained saliency modules.
- The RCM and MAM modules are complementary: RCM sharpens boundaries (precision), while MAM expands object coverage (recall), leading to synergistic performance gains.
- A momentum contrastive learning approach with EMA for the SupportNet provides more stable and reliable self-supervision signals compared to direct supervision or weight-sharing.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.