Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
WAFT-Stereo: Warping-Alone Field Transforms for Stereo Matching
Yihan Wang, Jia Deng
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 classification step before regression (vs regression only) (ablation) | WAFT-Stereo · ETH3D / Middlebury-Q train | BP-1-all / BP-2-all: 2.59 -> 1.48 ETH3D; 6.45 -> 4.62 Middlebury-Q | Improved |
| add regression refinement after classification (vs classification only) (ablation) | WAFT-Stereo · ETH3D / Middlebury-Q train | BP-1-all: 66.8 -> 1.48 | Improved |
| use cost volume for the classification step (ablation) | WAFT-Stereo · ETH3D / Middlebury-Q train | BP-1-all: 1.48 -> 1.53 | No clear effect |
| use 40 bins instead of 5 (ablation) | WAFT-Stereo · ETH3D train | BP-1-all: 2.46 -> 1.48 | Improved |
| use 80 bins instead of 40 (ablation) | WAFT-Stereo · ETH3D / Middlebury-Q train | BP-1-all / BP-2-all: BP-1-all 1.48 -> 1.27; BP-2-all 4.62 -> 5.39 | Mixed |
| add high-resolution ResNet blocks (ablation) | WAFT-Stereo · ETH3D train | BP-1-all: 1.99 -> 1.48 | Improved |
| use MoL loss instead of L1 (ablation) | WAFT-Stereo · ETH3D train | BP-1-all: 2.44 -> 1.48 | Improved |
| scale backbone from DAv2-S to DAv2-L (ablation) | WAFT-Stereo · ETH3D train | BP-1-all: 1.48 -> 0.80 | Improved |
| use DINOv3-L instead of DAv2-L (ablation) | WAFT-Stereo · ETH3D train | BP-1-all: 0.80 -> 0.87 | No clear effect |
| train on SceneFlow only instead of 7.7M synthetic pairs (ablation) | WAFT-Stereo (DAv2-S) · ETH3D train | BP-1-all: 0.86 -> 71.0 | Hurt |
| replace cost volumes with warping (full model) vs FoundationStereo | WAFT-Stereo · ETH3D test zero-shot | BP-0.5-noc: 2.31 -> 0.89 | Improved |
| WAFT-Stereo vs S2M2-XL on Middlebury BP-2 | WAFT-Stereo · Middlebury test | BP-2-noc: 1.15 vs 2.53 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FoundationStereo — FoundationStereo: Zero-Shot Stereo Matching (CVPR 2025) · better on 3 of 3 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
- Mono2Stereo: A Benchmark and Empirical Study for Stereo Conversion (CVPR 2025 · similarity 0.86)
- MonSter: Marry Monodepth to Stereo Unleashes Power (CVPR 2025 · similarity 0.85)
- Stereo Anywhere: Robust Zero-Shot Deep Stereo Matching Even Where Either Stereo or Mono Fail (CVPR 2025 · similarity 0.84)
- Any-Stereo: Arbitrary Scale Disparity Estimation for Iterative Stereo Matching (AAAI 2024 · similarity 0.83)
- DEFOM-Stereo: Depth Foundation Model Based Stereo Matching (CVPR 2025 · similarity 0.83)
- LoS: Local Structure-Guided Stereo Matching (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.
