Paper 2603.03269
LoGeR: Long-Context Geometric Reconstruction with Hybrid Memory
- Published
- Mar 2026
- Research lab
- Google DeepMind
- Citations
- 22
- GitHub
- 609 stars
01 In brief
Summary
LoGeR (Long-context Geometric Reconstruction) is a novel architecture for scaling feedforward dense 3D reconstruction to extremely long video sequences (up to 19k frames) without post-optimization.
It processes video in chunks, using bidirectional attention for intra-chunk reasoning and a hybrid memory module for inter-chunk coherence.
The hybrid memory combines a parametric Test-Time Training (TTT) memory to anchor the global coordinate frame and prevent scale drift, with a non-parametric Sliding Window Attention (SWA) mechanism to preserve uncompressed local context for precise adjacent alignment.
This design allows training on 128-frame sequences and generalization to thousands of frames.
Evaluated on KITTI and a repurposed VBR benchmark, LoGeR reduces Absolute Trajectory Error (ATE) on KITTI by over 74% (from 72.86 to 18.65) and achieves a 55.2% relative improvement on VBR sequences.
A variant, LoGeR*, incorporates a feedforward pose alignment step.
The method also outperforms prior work on short-sequence benchmarks (7-Scenes, ScanNet, TUM-Dynamics).
Ablations confirm the necessity of both memory components, a diverse large-scale training data mixture, and a progressive curriculum training strategy.
02 From the paper
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.