Reproduce “PhysInsert: Physically Plausible Object Insertion with Geometric Priors and Bidirectional Object-Scene Interactions” 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

Yue Gong, Le Zhang, Rui Chen, Song Chang, Xiaochao Qu, Luoqi Liu, Ting Liu, Yebin Liu

PhysInsert learns reference-based object insertion from paired scenes rendered with and without an object, combining scene geometry with supervision for the object’s effects on its surroundings. On 400 text-guided benchmark cases, it achieves 62.99% DIoU, versus 48.32% for Qwen-Image-Edit-2511, indicating better placement under that metric, not direct physical-correctness measurement. Identity preservation remains limited when viewpoints differ substantially.

Summary by alphaXiv.

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