Research map / 3D vision / Depth and stereo
NeRF-Supervised Deep Stereo
Fabio Tosi, Alessio Tonioni, Daniele De Gregorio, Matteo Poggi
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 |
|---|---|---|---|
| Train with NeRF-supervised triplets and filtered rendered disparity rather than the MfS pipeline | RAFT-Stereo · Midd-A F | F (>2px): 24.61 -> 13.12 | Improved |
| Train with NeRF-supervised triplets rather than the MfS pipeline | CFNet · KITTI-12 | D1 (>3px): 4.47 -> 4.64 | Hurt |
| Train using NeRF supervision rather than synthetic ground-truth training | RAFT-Stereo · ETH3D | All (>1px): 2.59 -> 2.94 | Mixed |
| Add rendered disparity supervision to triplet photometric supervision (ablation) | RAFT-Stereo · KITTI-12 | D1 (>3px): 5.11 -> 4.21 | Improved |
| Use triplet rather than pairwise photometric supervision alongside filtered disparity loss (ablation) | RAFT-Stereo · Midd-A F | F (>2px): 16.11 -> 15.31 | Improved |
| Filter NeRF-rendered disparity labels using the AO threshold (ablation) | RAFT-Stereo with rendered disparity supervision · KITTI-12 | D1 (>3px): 5.79 -> 4.49 | Improved |
| Add triplet photometric loss to filtered rendered disparity supervision (ablation) | RAFT-Stereo · Midd-A F | F (>2px): 15.50 -> 15.31 | Improved |
| Use rendered disparity supervision alone instead of triplet photometric supervision (ablation) | RAFT-Stereo · Midd-A F | F (>2px): 28.33 -> 21.71 | Hurt |
| Add AO-based weighting to the two loss terms (ablation) | RAFT-Stereo with triplet photometric and filtered disparity losses · Midd-21 | (>2px): 16.26 -> 14.87 | Improved |
| Train for 200K rather than 100K steps (ablation) | RAFT-Stereo with full NS loss · Midd-21 | (>2px): 14.87 -> 12.87 | Improved |
| Add medium and small baselines to the large baseline (ablation) | RAFT-Stereo · Midd-A F | F (>2px): 18.71 -> 14.92 | Improved |
| Add medium and small baselines to the large baseline (ablation) | RAFT-Stereo · KITTI-12 | D1 (>3px): 3.97 -> 4.31 | Hurt |
Closest papers by meaning
- Learning Stereo from Single Images (ECCV 2020 · similarity 0.91)
- Reversing the cycle: self-supervised deep stereo through enhanced monocular distillation (ECCV 2020 · similarity 0.90)
- Depth-Supervised NeRF: Fewer Views and Faster Training for Free (CVPR 2022 · similarity 0.90)
- Unsupervised Adaptation for Deep Stereo (ICCV 2017 · similarity 0.90)
- Zoom and Learn: Generalizing Deep Stereo Matching to Novel Domains (CVPR 2018 · similarity 0.89)
- Stereo Risk: A Continuous Modeling Approach to Stereo Matching (ICML 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.
