Research map / 3D vision / Depth and stereo
PixVOD: Pixel-Distributed Direct Visual Odometry and Depth Estimation
Shinjeong Kim, Ignacio Alzugaray, Callum Rhodes, Paul Kelly, Andrew Davison
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 |
|---|---|---|---|
| use hierarchical quad-tree identity factors instead of grid (ablation) | PixVOD · Replica | GBP iterations / FLOPs: 1711 FLOPs/px vs 1135 FLOPs/px; grid needs at least 50 times more GBP iterations | Improved |
| add photometric factors to normal integration (vs BINI alone) | PixVOD · Replica depth | depth quality | Improved |
| use DSINE normals instead of ground-truth normals (ablation) | PixVOD · Replica | pose error: 7.29 -> 7.21 RPE-r; 91.3 -> 99.3 RPE-t; 97 -> 93 success | Mixed |
| use ground-truth depth prior instead of estimated normals (ablation) | PixVOD · Replica | convergence | Improved |
| use keyframe tracking with frequent target updates (framerate 10 instead of 0.2) (ablation) | PixVOD · Replica | relative pose error | Improved |
| increase target-frame update rate from framerate 5 to 10 (ablation) | PixVOD · Replica | relative pose error | No clear effect |
| use mild photometric noise (ablation) | PixVOD · Replica with noise | convergence basin | Hurt |
| use half image resolution (ablation) | PixVOD · Replica | pose error | No clear effect |
| use quarter image resolution (ablation) | PixVOD · Replica | pose error | Hurt |
| use too weak or too strong camera motion identity factor (ablation) | PixVOD · Replica | pose error | Hurt |
| use identical factor strengths across layers instead of halving per layer (ablation) | PixVOD · Replica | convergence | Improved |
| use larger numbers of iterations per frame (ablation) | PixVOD · Replica | pose accuracy | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- DROID-SLAM — DROID-SLAM: Deep Visual SLAM for Monocular, Stereo, and RGB-D Cameras (NeurIPS 2021) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Direct Sparse Odometry (TPAMI 2017 · similarity 0.88)
- Unsupervised Learning of Monocular Depth Estimation and Visual Odometry With Deep Feature Reconstruction (CVPR 2018 · similarity 0.88)
- VOLDOR: Visual Odometry From Log-Logistic Dense Optical Flow Residuals (CVPR 2020 · similarity 0.88)
- Visual Odometry for Pixel Processor Arrays (ICCV 2017 · similarity 0.87)
- Leveraging Consistent Spatio-Temporal Correspondence for Robust Visual Odometry (AAAI 2025 · similarity 0.85)
- Depth2Pose: A Pose-Based Benchmark for Monocular Depth Estimation without Ground-Truth Depth (NeurIPS 2026 · similarity 0.85)
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.
