Research map / 3D vision / Depth and stereo
Revisit Self-supervision with Local Structure-from-Motion
Shengjie Zhu, Xiaoming Liu
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 |
|---|---|---|---|
| run local SfM self-supervision on 5 frames over supervised depth estimator output | ZoeDepth · ScanNet | SIlog: 6.655 -> 5.901 | Improved |
| run local SfM self-supervision on Metric3D output | Metric3D · ScanNet | SIlog: 6.708 -> 4.170 | Improved |
| run local SfM on KITTI360 | ZoeDepth · KITTI360 | SIlog: 14.154 -> 13.220 | Improved |
| align support frames with estimated depth adjustment scalars | ZoeDepth · ScanNet support frames | A.Rel: 0.104 -> 0.079 | Improved |
| refine correspondences with local SfM poses (RGB-D input) | PDC-Net · ScanNet correspondence | PCK-1: 0.119 -> 0.178 | Improved |
| refine correspondences of RoMa | RoMa · ScanNet correspondence | AEPE: 3.333 -> 3.067 | Improved |
| use prior LightedDepth refinement instead of the proposed method on correspondence | PDC-Net · ScanNet correspondence | PCK-1: 0.119 -> 0.061 | Hurt |
| use bundle-RANSAC-adjustment poses instead of COLMAP | sparse-view pose · ScanNet 5 frames | success rate (%): 36.7 -> 100.0 | Improved |
| use proposed pose initialization instead of SPARF NeRF joint optimization | pose estimation · Replica 3 frames | Rot.: 0.15 -> 0.52 | Hurt |
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.
- SPARF — SPARF: Neural Radiance Fields From Sparse and Noisy Poses (CVPR 2023) · better on 1 of 2 shared results
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
- Digging Into Self-Supervised Monocular Depth Estimation (ICCV 2019 · similarity 0.88)
- Self-Supervised Learning With Geometric Constraints in Monocular Video: Connecting Flow, Depth, and Camera (ICCV 2019 · similarity 0.88)
- Supervising the new with the old: learning SFM from SFM (ECCV 2018 · similarity 0.88)
- The Temporal Opportunist: Self-Supervised Multi-Frame Monocular Depth (CVPR 2021 · similarity 0.87)
- Learning to Discriminate Scene Structures Makes Self-Supervised Depth Learning Scalable (NeurIPS 2026 · similarity 0.87)
- Mining Supervision for Dynamic Regions in Self-Supervised Monocular Depth Estimation (CVPR 2024 · similarity 0.87)
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.
