Physically Consistent Object-Level Scene Reconstruction
Example result by CIR-phys [] (right). Left: ground truth. Middle: baseline deep learning-based approach.
In our work CIR-phys [] we include object interpenetration constraints into deep learning based object pose estimation (Lipson et al., Coupled Iterative Refinement for 6D Multi-Object Pose Estimation, CVPR 2022) to reduce interpenetrations of detected objects and improve the physical plausibility of the scene reconstruction.