Articulated Object Reconstruction from Rest-State Observation
The paper claims it can infer an articulated object’s parts and joints from one closed, rest-state view.
The method uses an explicit mesh to check and fuse outputs from vision-language and segmentation models. Because no motion is observed, it synthesizes possible articulations with a video diffusion model, then filters them by geometric consistency. The authors say it produces accurate part decomposition and physically plausible motion, competitive with baselines that do see motion. HF Daily Papers' note
The method uses an explicit mesh to check and fuse outputs from vision-language and segmentation models. Because no motion is observed, it synthesizes possible articulations with a video diffusion model, then filters them by geometric consistency. The authors say it produces accurate part decomposition and physically plausible motion, competitive with baselines that do see motion. HF Daily Papers' note
score 4