Research map / 3D vision / Depth and stereo
On the real-world generalisability of Optical Flow models
Petter Reijalt, Alexander Gielisse, Rickard Karlsson, Jan van Gemert
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 |
|---|---|---|---|
| restrict to models that beat RAFT on KITTI/Sintel/Spring (ablation) | Things-checkpoint optical flow models · real-world benchmarks (FlowFactor, Slow Flow, TAP-Flow) | Fl-all vs benchmark error: stagnate or even perform worse on real-world data | Hurt |
| add more synthetic pre-training data beyond C+T (ablation) | optical flow models · FlowFactor | Fl-all: does not necessarily improve performance | No clear effect |
| small amount of realistic data added (C+T+S+K+H) (ablation) | optical flow models · FlowFactor | Fl-all: highest accuracy with C+T or C+T+S+K+H | Improved |
| lighting-category accuracy as predictor of Slow Flow accuracy | FlowFactor · Slow Flow | Spearman correlation: rho = 0.75 | Improved |
| large-displacement accuracy as predictor of Slow Flow accuracy | FlowFactor · Slow Flow | Spearman correlation: rho = 0.74 | Improved |
| optimise for large displacement and repetitive textures among top models | top-10 models · FlowFactor lighting and occlusion | Spearman correlation: negative correlation between categories | Hurt |
| use 5 sparse points per pair instead of other point set | FlowFactor annotation · TAP-Vid-DAVIS 2 point sets | Spearman correlation: 0.95 | Improved |
| GMA global motion aggregation vs SEA-RAFT small under lighting change | GMA · FlowFactor lighting | qualitative flow | Improved |
Closest papers by meaning
- Shape and Texture: What Influences Reliable Optical Flow Estimation? (CVPR 2025 · similarity 0.89)
- Taming generative video models for zero-shot optical flow extraction (NeurIPS 2025 · similarity 0.89)
- Zero-Shot Monocular Scene Flow Estimation in the Wild (CVPR 2025 · similarity 0.88)
- StreamFlow: Streamlined Multi-Frame Optical Flow Estimation for Video Sequences (NeurIPS 2024 · similarity 0.88)
- DPFlow: Adaptive Optical Flow Estimation with a Dual-Pyramid Framework (CVPR 2025 · similarity 0.88)
- MemFlow: Optical Flow Estimation and Prediction with Memory (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.
