Research map / 3D vision / Depth and stereo
Revisiting Stereo Depth Estimation From a Sequence-to-Sequence Perspective With Transformers
Zhaoshuo Li, Xingtong Liu, Nathan Drenkow, Andy Ding, Francis X. Creighton, Russell H. Taylor, Mathias Unberath
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 attention mask (ablation) | STTR · Scene Flow | 3 px error: 3.61 -> 2.77 | Improved |
| add optimal transport (uniqueness constraint) (ablation) | STTR (AM) · Scene Flow | 3 px error: 2.77 -> 2.32 | Improved |
| add optimal transport (uniqueness constraint) (ablation) | STTR (AM) · Scene Flow | occlusion IOU: 0.77 -> 0.87 | Improved |
| add context adjustment layer (ablation) | STTR (AM+OT) · Scene Flow | 3 px error: 2.32 -> 2.21 | Improved |
| add context adjustment layer (ablation) | STTR (AM+OT) · Scene Flow | EPE: 0.70 -> 0.63 | Improved |
| add relative positional encoding (ablation) | STTR (AM+OT+CAL) · Scene Flow | 3 px error: 2.21 -> 1.26 | Improved |
| add relative positional encoding (ablation) | STTR (AM+OT+CAL) · Scene Flow | EPE: 0.63 -> 0.45 | Improved |
| increase attention stride from 3 to 5 at inference (ablation) | STTR · Scene Flow | 3 px error: 1.26 -> 1.70 | Hurt |
| increase attention stride from 3 to 4 at inference (ablation) | STTR · Scene Flow | EPE: 0.45 -> 0.71 | Hurt |
| train with asymmetric augmentation | PSMNet · KITTI 2015 (no fine-tuning) | 3 px error: 27.79 -> 7.43 | Improved |
| train with asymmetric augmentation | PSMNet · MPI Sintel (no fine-tuning) | 3 px error: 6.81 -> 7.93 | Hurt |
| use unlimited disparity (STTR) vs D=192 for prior work, all pixels | PSMNet · Scene Flow | 3 px error: 3.31 -> 1.26 | Improved |
Closest papers by meaning
- S2M2: Scalable Stereo Matching Model for Reliable Depth Estimation (ICCV 2025 · similarity 0.92)
- A Survey on Deep Learning Techniques for Stereo-Based Depth Estimation (TPAMI 2020 · similarity 0.89)
- Single View Stereo Matching (CVPR 2018 · similarity 0.89)
- Stereo Anywhere: Robust Zero-Shot Deep Stereo Matching Even Where Either Stereo or Mono Fail (CVPR 2025 · similarity 0.88)
- MonSter: Marry Monodepth to Stereo Unleashes Power (CVPR 2025 · similarity 0.87)
- RRT-MVS: Recurrent Regularization Transformer for Multi-View Stereo (AAAI 2025 · 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.
