Research map / Architectures and efficiency / Attention and transformers
Beyond Tracking: Selecting Memory and Refining Poses for Deep Visual Odometry
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Add adaptive Memory and pose Refining to tracking (ablation) | Ours · TUM-RGBD fr3/nstr_ntex_near_loop | translational RMSE (m/s): 0.371 -> 0.123 | Helped |
| Add temporal attention over stored memories (ablation) | Ours · TUM-RGBD fr2/pioneer_360 | translational RMSE (m/s): 0.1338 -> 0.056 | Helped |
| Add spatial attention over memory feature channels (ablation) | Ours · TUM-RGBD fr2/pioneer_360 | translational RMSE (m/s): 0.076 -> 0.056 | Helped |
| Add temporal attention over stored memories (ablation) | Ours · TUM-RGBD fr3/nstr_ntex_near_loop | translational RMSE (m/s): 0.195 -> 0.123 | Helped |
| Add spatial attention over memory feature channels (ablation) | Ours · TUM-RGBD fr3/nstr_ntex_near_loop | translational RMSE (m/s): 0.157 -> 0.123 | Helped |
| Use 11-frame rather than 5-frame sequences (ablation) | Ours · KITTI (Avg) | translational RMSE drift (%): 4.84 -> 3.47 | Helped |
| Use 11-frame rather than 5-frame sequences (ablation) | Ours · KITTI (Avg) | rotational RMSE drift (°/100m): 2.65 -> 1.75 | Helped |
| Use longer input sequences (5 to 11 frames) (ablation) | Ours · TUM-RGBD | translational RMSE (m/s): 0.230 -> 0.153; 0.106 -> 0.056; 0.201 -> 0.172 | Mixed |
| Use Memory and Refining to maintain global context and update previous poses | Ours · KITTI Seq 03, 04, 05, 06, 07, 10 (Avg) | translational RMSE drift (%): 4.97 -> 3.47 | Helped |
| Use Memory and Refining to maintain global context and update previous poses | Ours · KITTI Seq 03, 04, 05, 06, 07, 10 (Avg) | rotational RMSE drift (°/100m): 2.26 -> 1.75 | Helped |
| Use learned Memory and Refining rather than ORB-SLAM2 local mapping with loop closure | Ours · KITTI (Avg) | translational RMSE drift (%): 2.62 -> 3.47 | Hurt |
Closest papers by meaning
- LEAP-VO: Long-term Effective Any Point Tracking for Visual Odometry (CVPR 2024 · similarity 0.89)
- Adaptive VIO: Deep Visual-Inertial Odometry with Online Continual Learning (CVPR 2024 · similarity 0.87)
- Leveraging Consistent Spatio-Temporal Correspondence for Robust Visual Odometry (AAAI 2025 · similarity 0.86)
- MambaVO: Deep Visual Odometry Based on Sequential Matching Refinement and Training Smoothing (CVPR 2025 · similarity 0.84)
- ZeroVO: Visual Odometry with Minimal Assumptions (CVPR 2025 · similarity 0.84)
- From Variance to Veracity: Unbundling and Mitigating Gradient Variance in Differentiable Bundle Adjustment Layers (CVPR 2024 · similarity 0.83)
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.