Research map / Label-efficient and robust learning / Recognition and long-tail
Rethinking Optical Flow From Geometric Matching Consistent Perspective
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Pre-train the feature matching encoder on geometric image matching (ablation) | MatchFlow(G) · Sintel train clean | AEPE: 1.27 -> 1.03 | Helped |
| Pre-train the feature matching encoder on geometric image matching (ablation) | MatchFlow(G) · Sintel train final | AEPE: 2.84 -> 2.45 | Helped |
| Pre-train the feature matching encoder on geometric image matching (ablation) | MatchFlow(G) · KITTI-15 train | Fl-all: 14.4 -> 15.6 | Hurt |
| Use 8 QuadTree attention blocks rather than 0 (ablation) | MatchFlow(G) · Sintel train clean | AEPE: 1.36 -> 1.03 | Helped |
| Use 8 QuadTree attention blocks rather than 4 (ablation) | MatchFlow(G) · Sintel train clean | AEPE: 1.18 -> 1.03 | Helped |
| Use QuadTree attention instead of Linear attention (ablation) | Feature Matching Extractor · Sintel train final | AEPE: 3.00 -> 2.45 | Helped |
| Use QuadTree attention instead of GLA (ablation) | Feature Matching Extractor · Sintel train final | AEPE: 2.82 -> 2.45 | Helped |
| Add match initialization (ablation) | MatchFlow(G) · Sintel train clean | AEPE: 1.03 -> 1.44 | Hurt |
| Add matching loss (ablation) | MatchFlow(G) · KITTI-15 train | Fl-epe: 4.08 -> 4.78 | Hurt |
| Apply tiling to zero-padded KITTI images (ablation) | MatchFlow(G) · KITTI-15 train | Fl-all: 18.2 -> 16.8 | Helped |
| Resize KITTI images and apply tiling instead of resizing alone (ablation) | MatchFlow(G) · KITTI-15 train | Fl-all: 16.3 -> 15.6 | Helped |
| Use multi-scale training (ablation) | MatchFlow(G) · KITTI-15 train | Fl-all: 15.6 -> 15.2 | Mixed |
Closest papers by meaning
- Global Matching With Overlapping Attention for Optical Flow Estimation (CVPR 2022 · similarity 0.90)
- GMFlow: Learning Optical Flow via Global Matching (CVPR 2022 · similarity 0.90)
- Displacement-Invariant Matching Cost Learning for Accurate Optical Flow Estimation (NeurIPS 2020 · similarity 0.89)
- Optical Flow Matching: Reframing Optical Flow as Continuous Transport Dynamics (CVPR 2026 · similarity 0.89)
- UFM: A Simple Path towards Unified Dense Correspondence with Flow (NeurIPS 2025 · similarity 0.87)
- GIM: Learning Generalizable Image Matcher From Internet Videos (ICLR 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.