PhysInsert Enables Physically Plausible Object Insertion
Yue Gong and seven coauthors presented PhysInsert, a framework for reference-based object insertion that jointly adapts an inserted object and its surrounding scene. It uses geometric priors, a VGGT geometry stream, and 2D object-scene spatial supervision, alongside the Blender-rendered Insert World dataset and the PhysInsert-Bench benchmark. The authors report improved image fidelity on PhysInsert-Bench and DreamBooth, while a user study found more accurate placement and better lighting integration, perspective, and pose plausibility.
The work addresses limitations in reference-based insertion by jointly modeling object adaptation and environmental changes, rather than limiting the method to copying an object's appearance. Its dataset and benchmark include cross-environment and cross-view cases focused on illumination and pose adaptation.
Computer Science > Computer Vision and Pattern Recognition
arXiv:2610.40453 (cs)
[Submitted on 29 Oct 2026]
Title:PhysInsert: Physically Plausible Object Insertion with Geometric Priors and Bidirectional Object-Scene Interactions
Authors:Yue Gong, Le Zhang, Rui Chen, Song Chang, Xiaochao Qu, Luoqi Liu, Ting Liu, Yebin Liu
View a PDF of the paper titled PhysInsert: Physically Plausible Object Insertion with Geometric Priors and Bidirectional Object-Scene Interactions, by Yue Gong and 7 other authors
#PhysInsert#Insert World#PhysInsert-Bench#VGGT#Geometry-Guided Adaptation