Research map / 3D vision / Depth and stereo
Weakly Supervised Learning of Deep Metrics for Stereo Reconstruction
Stepan Tulyakov, Anton Ivanov, Francois Fleuret
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 |
|---|---|---|---|
| add uniqueness constraint (CONTRASTIVE) over MIL (ablation) | semi-supervised stereo metric · KITTI'12 | WTA error (%): 18.45 -> 17.63 | Improved |
| combine MIL and CONTRASTIVE (ablation) | semi-supervised stereo metric · KITTI'12 | WTA error (%): 17.63 -> 16.12 | Improved |
| add dynamic programming with continuity and ordering constraints (CONTRASTIVE-DP) (ablation) | CONTRASTIVE · KITTI'12 | WTA error (%): 17.63 -> 14.61 | Improved |
| use all constraints (CONTRASTIVE-DP) instead of only epipolar and disparity (MIL) (ablation) | MIL · KITTI'12 | WTA error (%): 18.45 -> 14.61 | Improved |
| DP training takes longer than CONTRASTIVE (ablation) | CONTRASTIVE-DP · KITTI'12 | training time (hr): 30 -> 68 | Hurt |
| replace supervised training with semi-supervised MC-CNN-SS | MC-CNN fst · KITTI'12 | WTA error (%): 15.44 -> 13.90 | Improved |
| replace supervised training with semi-supervised MC-CNN-SS | MC-CNN fst · KITTI'15 | WTA error (%): 15.38 -> 14.08 | Improved |
| replace supervised training with semi-supervised MC-CNN-SS | MC-CNN fst · Middlebury | WTA error (%): 29.94 -> 30.06 | No clear effect |
| replace supervised training with semi-supervised MC-CNN-SS in full pipeline | MC-CNN fst · KITTI'12 benchmark | pipeline error (%): 2.82 -> 3.02 | Hurt |
| replace supervised training with semi-supervised MC-CNN-SS in full pipeline | MC-CNN fst · KITTI'15 benchmark | pipeline error (%): 4.62 -> 4.97 | Hurt |
| replace supervised training with semi-supervised MC-CNN-SS in full pipeline | MC-CNN fst · Middlebury benchmark | pipeline error (%): 9.69 -> 12.3 | Hurt |
| train metric on the same dataset as the test set instead of another dataset (ablation) | MC-CNN-SS · KITTI'12 WTA | WTA error (%): 15.52 -> 13.90 | Improved |
Closest papers by meaning
- MetricAnything: Scaling Metric Depth Pretraining with Noisy Heterogeneous Sources (ECCV 2026 · similarity 0.89)
- End-To-End Learning of Geometry and Context for Deep Stereo Regression (ICCV 2017 · similarity 0.88)
- Degradation-Agnostic Correspondence From Resolution-Asymmetric Stereo (CVPR 2022 · similarity 0.88)
- NeRF-Supervised Deep Stereo (CVPR 2023 · similarity 0.87)
- Zoom and Learn: Generalizing Deep Stereo Matching to Novel Domains (CVPR 2018 · similarity 0.87)
- Reversing the cycle: self-supervised deep stereo through enhanced monocular distillation (ECCV 2020 · 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.
