Reproduce “Open World Object Detection in the Era of Foundation Models” 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
Orr Zohar, Alejandro Lozano, Shelly Goel, Serena Yeung, Kuan-Chieh Wang
Research from Stanford University redefines Open World Object Detection for the era of foundation models, introducing the challenging RWD benchmark and a novel attribute-based method, FOMO. FOMO achieved a 3x increase in unknown object mAP and a 20% improvement in known object mAP compared to baselines on the new benchmark.
Key insights
- Foundation models quickly saturate existing OWD benchmarks, achieving twice the performance of traditional methods, which indicates these benchmarks are no longer challenging enough for current models.
- Truly unknown objects in real-world scenarios can be effectively identified by leveraging their shared visual and functional attributes with known classes, a concept inspired by cognitive psychology.
- The proposed RWD benchmark successfully presents out-of-distribution challenges that expose the limitations of zero-shot foundation model baselines.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.