Reproduce “Follow Anything: Open-set detection, tracking, and following in real-time” 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
Alaa Maalouf, Ninad Jadhav, Krishna Murthy Jatavallabhula, Makram Chahine, Daniel M.Vogt, Robert J. Wood, Antonio Torralba, Daniela Rus
Follow Anything (FAn) is a real-time robotic system developed at MIT and Harvard for open-set detection, tracking, and following of any object using multimodal queries. It achieves 6-20 frames per second on commodity hardware and demonstrates robust performance in autonomously tracking diverse objects like RC cars and drones in dynamic environments.
Key insights
- Real-time open-set object following on commodity hardware is achievable by carefully optimizing and integrating large foundation models.
- A DINO-based direct detection approach (DINO-SOLO) offers a substantial speed-accuracy trade-off, enabling high frame rates with good detection accuracy but less refined masks compared to SAM.
- Multimodal querying and robust re-detection strategies, including cross-trajectory feature comparison, are crucial for maintaining object tracking in dynamic and challenging environments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.