Reproduce “One Weird Trick to Improve Your Semi-Weakly Supervised Semantic Segmentation Model” 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
Wonho Bae, Junhyug Noh, Milad Jalali Asadabadi, Danica J. Sutherland
Prediction filtering uses an image classifier to exclude classes judged absent before selecting pixel segmentation labels, avoiding pixel-level pseudo-labels. The classifier supplies image-level classes, while the segmentation model locates them. On PASCAL VOC 2012, VGG-based DeeplabV1 reaches 67.4 mIoU versus 61.8 with image-level supervision for 9.1K images; both use CRF post-processing. A missed small or occluded class can be wrongly removed.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.