Research map / Probabilistic and scientific modeling / Molecules and proteins
Flow Fields: Dense Correspondence Fields for Highly Accurate Large Displacement Optical Flow Estimation
Christian Bailer, Bertram Taetz, Didier Stricker
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 scales to hierarchical matching (k=0 to k=3) (ablation) | Flow Fields correspondence · MPI-Sintel training subset (non-occluded) | percent flows within 3 pixels: 79.13% -> 89.20% | Improved |
| add scales to hierarchical matching (k=0 to k=3) (ablation) | Flow Fields correspondence · MPI-Sintel training subset (non-occluded) | EPE: 32.51 -> 6.04 | Improved |
| add scales to hierarchical matching (k=0 to k=3) (ablation) | Flow Fields + EpicFlow · MPI-Sintel training subset (non-occluded) | EPE after EpicFlow: 2.81 -> 2.04 | Improved |
| add median filtering of outliers to k=3 (ablation) | Flow Fields correspondence · MPI-Sintel training subset (non-occluded) | percent flows within 3 pixels: 89.20% -> 92.17% | Improved |
| add median filtering of outliers to k=3 (ablation) | Flow Fields correspondence · MPI-Sintel training subset (non-occluded) | EPE after EpicFlow: 2.04 -> 2.13 | Hurt |
| use Flow Fields matching instead of ANNF (ablation) | Correspondence fields · MPI-Sintel training subset (non-occluded) | percent flows within 3 pixels: 68.05% -> 89.20% | Improved |
| add multi-scale matching (k=3 vs k=0) (ablation) | Flow Fields · Middlebury training | EPE of correspondence: 12.33 -> 1.16 | Improved |
| add sub-scales (ablation) | Flow Fields+ · MPI-Sintel training | EPE: 2.438 -> 2.410 | Improved |
| add sub-scales (ablation) | Flow Fields+ · KITTI 2015 training | >3px failure rate: 21.36% -> 21.22% | Improved |
| use 2x instead of 1x consistency check (ablation) | Flow Fields+ Fast · MPI-Sintel training | EPE: 2.461 -> 2.448 | Improved |
| use Flow Fields+ Fast instead of full Flow Fields+ (ablation) | Flow Fields+ · MPI-Sintel training | runtime: 14.0s -> 6.7s | Improved |
| use Flow Fields+ over original Flow Fields (ablation) | Flow Fields · KITTI 2015 training | >3px failure rate: 24.74% -> 21.22% | Improved |
Closest papers by meaning
- Efficient Coarse-To-Fine PatchMatch for Large Displacement Optical Flow (CVPR 2016 · similarity 0.92)
- Robust Interpolation of Correspondences for Large Displacement Optical Flow (CVPR 2017 · similarity 0.91)
- SceneFlowFields++: Multi-frame Matching, Visibility Prediction, and Robust Interpolation for Scene Flow Estimation (IJCV 2019 · similarity 0.91)
- Fast Optical Flow Using Dense Inverse Search (ECCV 2016 · similarity 0.90)
- MegaFlow: Zero-Shot Large Displacement Optical Flow (ECCV 2026 · similarity 0.90)
- UFM: A Simple Path towards Unified Dense Correspondence with Flow (NeurIPS 2025 · similarity 0.89)
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.
