Research map / Video / Video generation
LoGeR: Long-Context Geometric Reconstruction with Hybrid Memory
Junyi Zhang, Charles Herrmann, Junhwa Hur, Chen Sun, Ming-Hsuan Yang, Forrester Cole, Trevor Darrell, Deqing Sun
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| remove TTT layers from the hybrid memory (ablation) | LoGeR · ScanNet subset 1000f | ATE: 0.162 -> 0.107 | Improved |
| remove SWA layers (ablation) | LoGeR · ScanNet subset 1000f | ATE: 0.143 -> 0.107 | Improved |
| add TTT on top of SWA (ablation) | LoGeR · TUM 1000f | ATE: 0.079 -> 0.050 | Improved |
| add SWA on top of TTT (ablation) | LoGeR · TUM 1000f | ATE: 0.053 -> 0.050 | Improved |
| add five large-scale datasets to the training mixture (ablation) | LoGeR · ScanNet subset 1000f | ATE: 0.156 -> 0.107 | Improved |
| add progressive curriculum training (ablation) | LoGeR · ScanNet subset 1000f | ATE: 0.133 -> 0.107 | Improved |
| add progressive curriculum training (ablation) | LoGeR* · TUM 500f | ATE: 0.029 -> 0.031 | Hurt |
| add feedforward SE(3) pose alignment (LoGeR* vs LoGeR) | LoGeR · KITTI avg | ATE (m): 25.44 -> 18.65 | Improved |
| add hybrid memory over chunk stitching (Pi3-Chunk -> LoGeR) | Pi3-Chunk · KITTI avg | ATE (m): 52.07 -> 25.44 | Improved |
| add hybrid memory over chunk stitching | Pi3-Chunk · VBR avg | ATE (m): 70.09 -> 36.16 | Improved |
| chunk stitching vs LoGeR on short horizons | LoGeR · VBR ~1k frames | ATE | No clear effect |
| smaller inference chunk size (64 -> 32) | LoGeR · 500-frame efficiency | speed / memory: 9.3 -> 12.1 FPS; 27.2 -> 18.1 GB | Mixed |
Closest papers by meaning
- Rethinking the State Update Gate for Long-Sequence Recurrent 3D Reconstruction (NeurIPS 2026 · similarity 0.89)
- Stable3R: Streaming 3D Reconstruction with Stable Geometric References (NeurIPS 2026 · similarity 0.89)
- OVGGT: O(1) Constant-Cost Streaming Visual Geometry Transformer (ECCV 2026 · similarity 0.89)
- FrameVGGT: Coherence-Preserving Memory for Bounded Streaming Geometry (NeurIPS 2026 · similarity 0.88)
- GHOST: Geometry-Hierarchical Online Streaming Token Eviction for Efficient 3D Reconstruction (NeurIPS 2026 · similarity 0.87)
- CLoSeR: Closing the Loop for Long-Context Streaming Reconstruction (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
