Research map / Architectures and efficiency / RNNs and normalization
GMFlow: Learning Optical Flow via Global Matching
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 |
|---|---|---|---|
| Replace local cost-volume regression with global feature matching using a Transformer and softmax | Optical flow estimation · Sintel train clean | EPE: 1.41 -> 1.08 | Helped |
| Use one GMFlow refinement rather than 31 RAFT refinements | Optical flow estimation · Sintel train clean | inference time (A100): 91 -> 66 | Helped |
| Use global matching instead of local matching (ablation) | GMFlow · Things val clean | EPE: 31.78 -> 6.67 | Helped |
| Add cross-attention between the two frame features (ablation) | GMFlow · Things val | EPE: 10.84 -> 6.67 | Helped |
| Add positional encoding (ablation) | GMFlow · Things val | EPE: 8.38 -> 6.67 | Helped |
| Add the feed-forward network (ablation) | GMFlow · Things val | EPE: 8.71 -> 6.67 | Helped |
| Add self-attention to Transformer feature enhancement (ablation) | GMFlow · Things val | EPE: 7.04 -> 6.67 | Helped |
| Add flow propagation using feature self-attention (ablation) | GMFlow · Sintel clean, unmatched pixels | EPE: 15.54 -> 10.39 | Helped |
| Add flow propagation using feature self-attention (ablation) | GMFlow · Sintel final, unmatched pixels | EPE: 19.50 -> 15.52 | Helped |
| Add higher-resolution refinement (ablation) | GMFlow · Sintel train clean | EPE: 1.65 -> 1.20 | Helped |
| Share Transformer weights between global matching and refinement (ablation) | GMFlow with refinement · Things clean | EPE: 3.41 -> 3.21 | Helped |
| Share multi-scale backbone features (ablation) | GMFlow with refinement · Sintel train clean | EPE: 1.27 -> 1.20 | Helped |
Closest papers by meaning
- Global Matching With Overlapping Attention for Optical Flow Estimation (CVPR 2022 · similarity 0.93)
- Gaussian Mixture Flow Matching Models (ICML 2025 · similarity 0.91)
- StreamFlow: Streamlined Multi-Frame Optical Flow Estimation for Video Sequences (NeurIPS 2024 · similarity 0.90)
- MemFlow: Optical Flow Estimation and Prediction with Memory (CVPR 2024 · similarity 0.89)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.89)
- EDCFlow: Exploring Temporally Dense Difference Maps for Event-based Optical Flow Estimation (CVPR 2025 · similarity 0.88)
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.