Research map / Probabilistic and scientific modeling / Molecules and proteins
GAFlow: Incorporating Gaussian Attention into Optical Flow
Ao Luo, Fan Yang, Xin Li, Lang Nie, Chunyu Lin, Haoqiang Fan, Shuaicheng Liu
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 Gaussian-Constrained Layer (GCL) to RAFT feature encoder (ablation) | RAFT · Sintel Val Clean | EPE: 1.43 -> 1.33 | Improved |
| use Gaussian-constrained kernel instead of learnable kernel without Gaussian (ablation) | GCL · Sintel Val Final | EPE: 2.76 -> 2.67 | Improved |
| use GCL instead of Swin Transformer block (ablation) | RAFT · Sintel Val Clean | EPE: 1.45 -> 1.33 | Improved |
| use GCL instead of cosFormer (ablation) | RAFT · Sintel Val Clean | EPE: 1.29 -> 1.33 (Clean); 2.82 -> 2.67 (Final) | Mixed |
| use single POLA block instead of GCL (ablation) | RAFT · Sintel Val Clean | EPE: 1.33 -> 1.42 | Hurt |
| add stronger POLA motion encoder (ablation) | GAFlow · Sintel Val Clean | EPE: 1.33 -> 1.18 | Improved |
| replace GGAC with deformable-kernel GGAD (ablation) | GAFlow · Sintel Val Clean | EPE: 1.13 -> 1.08 | Improved |
| add SKBlocks to the decoder (ablation) | GAFlow · KITTI-15 Val | EPE: 4.26 -> 3.98 | Improved |
| use local Gaussian attention instead of global attention (GMA) | GAFlow-S · Sintel train Final (C+T) | EPE: 2.74 -> 2.56 | Improved |
| apply GAFlow modules on top of FlowFormer | FlowFormer · KITTI-15 Val | F1-all: 14.7 -> 13.9 | Improved |
| GAFlow versus DIP on KITTI test | GAFlow · KITTI-15 Test | F1-all: 4.21 -> 4.52 | Hurt |
Later papers that compared with this method
1 later accepted paper reports GAFlow as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- FlowFM: Advancing Dark Optical Flow Estimation with Flow Matching (CVPR 2026 · better on 1 of 1 shared result)
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
- GMFlow: Learning Optical Flow via Global Matching (CVPR 2022 · similarity 0.89)
- Global Matching With Overlapping Attention for Optical Flow Estimation (CVPR 2022 · similarity 0.89)
- Rethinking Optical Flow From Geometric Matching Consistent Perspective (CVPR 2023 · similarity 0.88)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.88)
- StreamFlow: Streamlined Multi-Frame Optical Flow Estimation for Video Sequences (NeurIPS 2024 · similarity 0.87)
- Gaussian Mixture Flow Matching Models (ICML 2025 · 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.
