Reproduce “HEAP: Unsupervised Object Discovery and Localization with Contrastive Grouping” 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
Xin Zhang, Jinheng Xie, Yuan Yuan, Michael Bi Mi, Robby T. Tan
HEAP introduces a hierarchical framework for unsupervised object discovery and localization, leveraging contrastive grouping across patch, region, and image levels, both intra- and inter-image. It achieves state-of-the-art performance on various benchmarks, including a 3.3% mIoU improvement over FOUND for multi-object retrieval on VOC12.
Key insights
- Hierarchical understanding, incorporating patch, region, and image-level relationships, is essential for comprehensive object perception.
- Leveraging cross-image relationships through contrastive learning significantly enhances grouping consistency and object discriminability.
- Unsupervised loss functions can be designed to jointly enable coherent intra-image grouping, foreground-background disentanglement, and inter-image semantic consistency.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.