Research map / Architectures and efficiency / RNNs and normalization
Learning Optical Flow From a Few Matches
Check 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 |
|---|---|---|---|
| Use Sparse Correlation Volume with k=8 and 1/4-resolution features instead of RAFT's 1/8-resolution dense volume | Optical flow estimation · Sintel test Clean, two-view | EPE: 1.94 -> 1.72 | Helped |
| Use Sparse Correlation Volume with k=8 | Optical flow estimation · Sintel test Final, two-view | EPE: 3.18 -> 3.60 | Hurt |
| Use Sparse Correlation Volume with k=8 | Optical flow estimation · KITTI-15 test | F1-all: 5.10 -> 6.17 | Hurt |
| Use Sparse Correlation Volume with k=8 | Optical flow estimation · Sintel test Clean, warm-start | EPE: 1.61 -> 1.77 | Hurt |
| Increase feature resolution from 1/8 to 1/4 (ablation) | Sparse Correlation Volume · Sintel train Clean | EPE: 1.55 -> 1.29 | Helped |
| Increase feature resolution from 1/8 to 1/4 (ablation) | Sparse Correlation Volume · KITTI-15 train | EPE: 5.74 -> 6.80 | Hurt |
| Increase the retained matches from k=1 to k=8 (ablation) | Sparse Correlation Volume · Sintel train Clean | EPE: 1.97 -> 1.55 | Helped |
| Increase the retained matches from k=1 to k=8 (ablation) | Sparse Correlation Volume · KITTI-15 train | F1-all: 25.6 -> 20.2 | Helped |
| Increase the retained matches from k=8 to k=32 (ablation) | RAFT correlation volume · Sintel train Clean | EPE: 1.64 -> 1.50 | Helped |
| Increase the retained matches from k=8 to k=32 (ablation) | RAFT correlation volume · KITTI-15 train | F1-all: 22.8 -> 19.5 | Helped |
| Keep only positive correlations using ReLU rather than the dense correlation volume (ablation) | RAFT · KITTI-15 train | F1-all: 17.5 -> 16.7 | Helped |
| Use Sparse Correlation Volume rather than dense correlation volume | Optical flow training at 1/4 resolution · Sintel training memory, batch=1 | memory: 10.6 GB -> 6.1 GB | Helped |
Closest papers by meaning
- Efficient All-Pairs Correlation Volume Sampling for Optical Flow Estimation (CVPR 2026 · similarity 0.89)
- GMFlow: Learning Optical Flow via Global Matching (CVPR 2022 · similarity 0.88)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.88)
- Displacement-Invariant Matching Cost Learning for Accurate Optical Flow Estimation (NeurIPS 2020 · similarity 0.88)
- DPFlow: Adaptive Optical Flow Estimation with a Dual-Pyramid Framework (CVPR 2025 · similarity 0.86)
- EDCFlow: Exploring Temporally Dense Difference Maps for Event-based Optical Flow Estimation (CVPR 2025 · similarity 0.86)
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.