Reproduce “Semi-Supervised Semantic Segmentation Based on Pseudo-Labels: A Survey” 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

Lingyan Ran, Yali Li, Guoqiang Liang, Yanning Zhang

This survey paper from Northwestern Polytechnical University offers a systematic review and categorization of pseudo-label methods in semi-supervised semantic segmentation. It details advancements across model structure design, pseudo-label refinement techniques, and optimization strategies, and examines their applications in specialized domains like medical and remote sensing imagery.

Key insights

  • Pseudo-label methods are categorized into single-model-based (e.g., self-training iteration, auxiliary tasks) and mutual-model-based (e.g., cross-pseudo-supervision, dynamic multi-training) approaches.
  • Significant effort in pseudo-label research is directed towards enhancing pseudo-label quality through active correction, de-biasing, and sophisticated confidence-based filtering strategies.
  • Beyond architectural changes and label refinement, diverse optimization techniques, including improved loss functions and more effective utilization of limited labeled data, are crucial for robust training.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.