Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Learning Representations from Foundation Models for Domain Generalized Stereo Matching
Yongjian Zhang, Longguang Wang, Kunhong Li, WANG Yun, Yulan Guo
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 |
|---|---|---|---|
| integrate DINOv2/DAM foundation-model features into PSMNet via FormerStereo | PSMNet · ETH3D train (zero-shot) | Bad 1.0 (noc) error (%): 10.2 -> 6.7 | Improved |
| integrate foundation model into RAFT-Stereo | RAFT-Stereo · ETH3D train (zero-shot) | Bad 1.0 (noc) error (%): 3.3 -> 3.3 | No clear effect |
| use learnable feature convex upsampling instead of bilinear interpolation (ablation) | FormerStereo decoder · ETH3D (ablation, PSMNet) | Bad 1.0 (noc) error (%): 7.68 -> 6.73 | Improved |
| add image reconstruction constraint (ablation) | FormerStereo decoder · ETH3D (ablation, PSMNet) | Bad 1.0 (noc) error (%): 9.45 -> 6.73 | Improved |
| use C4 cost space instead of plain concatenation cost space (ablation) | FormerStereo · ETH3D (ablation, PSMNet) | Bad 1.0 (noc) error (%): 12.42 -> 6.73 | Improved |
| use C4 cost space instead of cosine cost space (ablation) | FormerStereo · KITTI 2015 (ablation, PSMNet) | Bad 3.0 (all) error (%): 6.22 -> 4.97 | Improved |
| use larger ViT encoder (Small to Large) (ablation) | FormerStereo · ETH3D (ablation, PSMNet) | Bad 1.0 (noc) error (%): 10.15 -> 6.73 | Improved |
| add asymmetric chromatic augmentation (ablation) | FormerStereo · KITTI 2015 (ablation, PSMNet) | Bad 3.0 (all) error (%): 5.75 -> 4.97 | Improved |
| swap DINOv2 for SAM or DAM as foundation model encoder (ablation) | FormerStereo · Middlebury half (ablation, PSMNet) | Bad 2.0 (noc) error (%): 7.72 -> 6.80 (DAM), 9.51 (SAM) | Mixed |
| use feature transformation module instead of linear projection on Swin encoder (ablation) | Swin-L encoder · ETH3D (ablation, PSMNet) | Bad 1.0 (noc) error (%): 5.84 -> 4.57 | Improved |
| use FormerStereo over Mask-PSMNet for stability across epochs | PSMNet · KITTI 2015 volatility (last 10 epochs) | Bad 3.0 error mean+-std (%): 6.4±0.31 -> 5.2±0.13 | Improved |
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.
- CFNet — CFNet: Cascade and Fused Cost Volume for Robust Stereo Matching (CVPR 2021) · better on 1 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
- GraftNet: Towards Domain Generalized Stereo Matching With a Broad-Spectrum and Task-Oriented Feature (CVPR 2022 · similarity 0.92)
- Domain Generalized Stereo Matching via Hierarchical Visual Transformation (CVPR 2023 · similarity 0.91)
- FoundationStereo: Zero-Shot Stereo Matching (CVPR 2025 · similarity 0.90)
- Hierarchical Object-Aware Dual-Level Contrastive Learning for Domain Generalized Stereo Matching (NeurIPS 2024 · similarity 0.90)
- All-in-One: Transferring Vision Foundation Models into Stereo Matching (AAAI 2025 · similarity 0.89)
- Robust Synthetic-to-Real Transfer for Stereo Matching (CVPR 2024 · similarity 0.88)
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.
