Research map / Video / Video segmentation and tracking
Triangular Consistency as a Universal Constraint for Learning Optical Flow
Yi Xiao, Carlos Rodriguez Coronel, Jing Zhan, Haniyeh Oskouie, Alex Wong, DONG LAO
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 triangular consistency as the only loss for one-epoch adaptation | RAFT pre-trained on FlyingChairs · Sintel Clean | EPE: 2.54 -> 2.08 | Improved |
| add triangular consistency as the only loss for one-epoch adaptation | RAFT pre-trained on FlyingThings3D · Sintel Final | EPE: 3.41 -> 3.13 | Improved |
| add temporal consistency during adaptation (vs augmentation-only or cycle-only) (ablation) | RAFT adaptation · Sintel Clean | EPE: 2.01 -> 1.77 | Improved |
| add augmentation consistency on top of temporal during adaptation (ablation) | RAFT adaptation · Sintel Clean | EPE: 1.77 -> 1.73 | Improved |
| augmentation-only or cycle-only losses for adaptation (ablation) | RAFT adaptation · Sintel | EPE | No clear effect |
| add triangular consistency to ARFlow trained on Sintel | ARFlow · Sintel Clean/Final test | EPE: 4.78 -> 4.48 (Clean), 5.89 -> 5.86 (Final) | Improved |
| add triangular consistency to ARFlow trained on KITTI | ARFlow · KITTI train Fl-all | Fl-all: 9.87 -> 9.94 | No clear effect |
| add triangular consistency to ARFlow trained on KITTI | ARFlow · HD1K zero-shot | EPE: 2.32 -> 1.99 | Improved |
| add augmentation consistency only to ARFlow (ablation) | ARFlow · Sintel Clean | EPE: 2.79 -> 2.60 | Improved |
| add cycle consistency only to ARFlow (ablation) | ARFlow · Sintel Final | EPE: 3.73 -> 3.69 (best), 3.77 (low weight) | Mixed |
| combine all three losses with weights 0.01/0.003/0.005 (ablation) | ARFlow · Sintel Clean/Final | EPE: 2.79/3.73 -> 2.58/3.49 | Improved |
| raise augmentation weight to 0.02 in full combination (ablation) | ARFlow · Sintel Clean | EPE: 2.58 -> 2.68 | 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.
- ARFlow — ARFlow: Auto-regressive Optical Flow Estimation for Arbitrary-Length Videos via Progressive Next-Frame Forecasting (ICLR 2026) · better on 5 of 6 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
- Imposing Consistency for Optical Flow Estimation (CVPR 2022 · similarity 0.92)
- Optical Flow Matching: Reframing Optical Flow as Continuous Transport Dynamics (CVPR 2026 · similarity 0.88)
- Consistency Guided Scene Flow Estimation (ECCV 2020 · similarity 0.88)
- Learning by Analogy: Reliable Supervision From Transformations for Unsupervised Optical Flow Estimation (CVPR 2020 · similarity 0.88)
- From Contrast to Consistency: Rethinking Event-based Continuous-Time Optical Flow Estimation (CVPR 2026 · similarity 0.87)
- OCAI: Improving Optical Flow Estimation by Occlusion and Consistency Aware Interpolation (CVPR 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.
