Real-world multi-camera captures often use only a few low-overlap views, leaving large areas poorly observed and causing noticeable artifacts in volumetric reconstructions. This work addresses that gap by separating background densification from human modeling: it leverages video diffusion to synthesize dense background supervision while using a robust deformable Gaussian human initialization and a recursive, motion-adaptive consistency injection to harmonize the composed result. The result is markedly improved novel-view synthesis and cleaner 4D outputs suitable for editing and re-rendering.
4D Human-Scene Reconstruction from Low-Overlap Captures
Reconstructs 4D dynamic human scenes from sparse, low-overlap multi-camera captures by decoupling background synthesis and human modeling. Synthesizes hundreds of camera-controlled background views with a video diffusion model, initializes deformable Gaussian humans via cross-view identity and triangulated keypoints, then applies motion-adaptive recursive enhancement to reduce artifacts.
Introduction
More Items
Continuously records egocentric visual and audio streams into a lightweight streaming memory that organizes experiences into current, short-term, and long-term tiers and retrieves multimodal evidence to answer queries about past events. Built for on-device use (smartphones/AI glasses) with dynamic retrieval routing.
AI Agent Papers2026
Unifies high-level visual-language reasoning and low-level control for visual navigation by decoupling cognition and control: a slow vision-language reasoner produces pixel goals with explicit chain-of-thought, and a fast action expert converts those anchors into continuous waypoints for robust urban and indoor navigation.
Explores unsupervised visual pretraining on visually rich documents to improve language-model intelligence; shows visual-pretrained models outperform text-only counterparts on the same corpora. Key aspects: direct use of images/layouts (no OCR-only pipeline), scalable across backbones and benchmarks.