Reproduce “HDVS: semi-supervised semantic segmentation via heterogeneous dual-branch voting supervision” 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
Yongqi Shan, Yunzhi Zhuge, Huchuan Lu
HDVS addresses costly pixel-level labeling in semantic segmentation by learning from labeled and unlabeled images. It pairs a CNN and a Transformer, using confidence-weighted randomized selection between their top pixel predictions to create pseudo-labels, then adds feature communication. On PASCAL VOC 2012’s 183-image labeled partition, HDVS reports 75.6% mIoU versus 74.7% for Diverse Co-training, a benchmark-specific validation gain.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.