LoGeR: Long-Context Geometric Reconstruction with Hybrid Memory

arXiv:2603.03269v2 Announce Type: replace Abstract: Feedforward geometric foundation models achieve strong short-window reconstruction, yet scaling them to minutes-long videos is bottlenecked by quadratic attention complexity or limited effective memory in recurrent designs. We present LoGeR (Long-context Geometric Reconstruction), a novel architecture that scales dense 3D reconstruction to extremely long sequences without post-optimization. LoGeR processes video streams in chunks, leveraging strong bidirectional priors for high-fidelity intra-chunk reasoning. To manage the critical challenge of coherence across chunk boundaries, we propose a learning-based hybrid memory module. This dual-component system combines a parametric Test-Time Training (TTT) memory to anchor the global coordinate frame and prevent scale drift, alongside a non-parametric Sliding Window Attention (SWA) mechanism to preserve uncompressed context for high-precision adjacent alignment. Remarkably, this memory architecture enables LoGeR to be trained on sequences of 128 frames, and generalize up to thousands of frames during inference. Evaluated across standard benchmarks and a newly repurposed VBR dataset with sequences of up to 19k frames, LoGeR substantially outperforms prior state-of-the-art feedforward methods--reducing ATE on KITTI by over 74%--and achieves robust, globally consistent reconstruction over unprecedented horizons.

Related

• MotionScape: A Large-Scale Real-World Highly Dynamic UAV Video Dataset for World Models
• Echoes Over Time: Unlocking Length Generalization in Video-to-Audio Generation Models
• VideoStir: Understanding Long Videos via Spatio-Temporally Structured and Intent-Aware RAG
• World-R1: Reinforcing 3D Constraints for Text-to-Video Generation
• Hybrid Latents -- Geometry-Appearance-Aware Surfel Splatting

Source: arXiv cs.CV | 2026-04-28