Research map / Architectures and efficiency / RNNs and normalization
RAFT: Recurrent All-Pairs Field Transforms for Optical Flow
Zachary Teed, Jia Deng
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 ConvGRU update operator instead of plain convolutions (ablation) | RAFT · Sintel train clean (C+T) | EPE: 2.04 -> 1.63 | Improved |
| add ConvGRU update operator instead of plain convolutions (ablation) | RAFT · KITTI-15 train (C+T) | F1-all (%): 26.1 -> 19.8 | Improved |
| tie weights across update iterations (ablation) | RAFT · Sintel train clean (C+T) | EPE: 1.96 -> 1.63 | Improved |
| tie weights across update iterations (ablation) | RAFT · KITTI-15 train (C+T) | F1-all (%): 24.1 -> 19.8 | Improved |
| add context network (ablation) | RAFT · Sintel train clean (C+T) | EPE: 1.93 -> 1.63 | Improved |
| add context network (ablation) | RAFT · KITTI-15 train (C+T) | F1-all (%): 23.1 -> 19.8 | Improved |
| use multi-scale features instead of single-scale (ablation) | RAFT · Sintel train clean (C+T) | EPE: 1.63 -> 2.08 | Hurt |
| increase lookup radius from 0 to 4 (ablation) | RAFT · Sintel train clean (C+T) | EPE: 3.41 -> 1.63 | Improved |
| add correlation pyramid pooling (ablation) | RAFT · Sintel train clean (C+T) | EPE: 1.95 -> 1.63 | Improved |
| use all-pairs correlation instead of local 32px range (ablation) | RAFT · Sintel train clean (C+T) | EPE: 2.91 -> 1.63 | Improved |
| use all-pairs correlation instead of local 128px range (ablation) | RAFT · Sintel train clean (C+T) | EPE: 1.64 -> 1.63 | No clear effect |
| use correlation lookup instead of warping for refinement (ablation) | RAFT · KITTI-15 train (C+T) | F1-epe: 11.83 -> 5.54 | Improved |
Later papers that compared with this method
14 later accepted papers report RAFT as a baseline in their results tables; 12 of them report a better number on most of the results they share with it.
- Efficient All-Pairs Correlation Volume Sampling for Optical Flow Estimation (CVPR 2026 · better on 1 of 1 shared result)
- Removing Cost Volumes from Optical Flow Estimators (ICCV 2025 · better on 1 of 1 shared result)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · better on 2 of 2 shared results)
- FlowNAS: Neural Architecture Search for Optical Flow Estimation (IJCV 2023 · better on 6 of 6 shared results)
- Unsupervised Optical Flow Estimation with Dynamic Timing Representation for Spike Camera (NeurIPS 2023 · better on 2 of 2 shared results)
- FlowFormer: A Transformer Architecture for Optical Flow (ECCV 2022 · better on 7 of 7 shared results)
- DIP: Deep Inverse Patchmatch for High-Resolution Optical Flow (CVPR 2022 · better on 4 of 6 shared results)
- Deep Equilibrium Optical Flow Estimation (CVPR 2022 · better on 4 of 5 shared results)
6 more not listed.
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MS-RAFT+: High Resolution Multi-Scale RAFT (IJCV 2023 · similarity 0.93)
- "SEA-RAFT: Simple, Efficient, Accurate RAFT for Optical Flow" (ECCV 2024 · similarity 0.89)
- PV-RAFT: Point-Voxel Correlation Fields for Scene Flow Estimation of Point Clouds (CVPR 2021 · similarity 0.88)
- Disentangling Architecture and Training for Optical Flow (ECCV 2022 · similarity 0.88)
- Context-Aware Iteration Policy Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.84)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.83)
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.
