Researchers have developed GEAR, a Geometry-Enabled Attention Routing framework designed for long-horizon video generation. This approach uses geometry as an explicit address for visual memory, enabling more efficient and accurate recovery of previously observed content. Unlike methods that reconstruct scenes into persistent 3D memory, GEAR uses per-frame geometry to establish token-level correspondences, avoiding accumulated geometric errors. The framework incorporates Geometric Correspondence Attention (GCA) to selectively inject relevant historical features and an Invisible Octree to filter out occluded correspondences, achieving state-of-the-art results in visual quality, camera control, and consistency for extended video generation. AI
IMPACT This framework could improve the efficiency and quality of long-horizon video generation by enhancing how models access and utilize visual memory.
RANK_REASON The cluster contains a research paper detailing a new framework for video generation. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
- Geometric Correspondence Attention
- Geometry as Address: Routing Attention to Visual Memory for Long-Horizon Camera-Controlled Video Generation
- Invisible Octree
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →