Unsupervised Motion Disambiguation via Semantic Scene Priors
A method that integrates pretrained semantic segmentation and depth-aware motion priors to resolve camera-object motion ambiguities without relying on heuristic background inpainting.
Source paper
Yifan Wang, Weiguang Zhao, Ni Yang, Haoyu Guo, Junting Dong, Weinan Zhang, Chunhua Shen
Generated by AI. The paper’s authors did not write these.
A method that integrates pretrained semantic segmentation and depth-aware motion priors to resolve camera-object motion ambiguities without relying on heuristic background inpainting.
A method that replaces absolute coordinate regression with relative graph-based embeddings to eliminate distance-from-origin accumulation errors in long-sequence reconstructions.
The paper introduces pi4, a unified foundation model for 4D scene perception that reconstructs dense geometry, camera poses, and tracks from uncalibrated RGB observations. By employing a shared DINOv3 encoder, looped Transformer decoders, and a geometry-conditioned motion branch, the model jointly achieves state-of-the-art multi-view scene reconstruction and 3D point tracking.
The model uses a shared DINOv3 image encoder feeding two independent task-specific looped Transformer decoders for geometry and motion. The geometry branch predicts dense local pointmaps, camera poses, and scale, which are then refined via a sparse 3D U-Net. The motion branch conditions on detached geometry features using spatial AdaLN to predict image-space correspondences that are analytically lifted into 3D using the reconstructed target geometry.