Research map / Video / Video segmentation and tracking
ARFlow: Auto-regressive Optical Flow Estimation for Arbitrary-Length Videos via Progressive Next-Frame Forecasting
Jiuming Liu, Mengmeng Liu, Siting Zhu, Yunpeng Zhang, Jiangtao Li, Michael Ying Yang, Francesco Nex, Hao Cheng, Hesheng Wang
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 Auto-regressive Flow Initialization (AFI) (ablation) | ARFlow · KITTI-15 train (C+T) | EPE: 3.05 -> 2.86 | Improved |
| add AFI (ablation) | ARFlow · Sintel train (C+T) | EPE Clean/Final: 0.95/2.15 -> 0.88/2.07 | Improved |
| add Auto-regressive Multi-stride Flow Refinement (AMFR) (ablation) | ARFlow · KITTI-15 train (C+T) | EPE: 3.00 -> 2.86 | Improved |
| use strides 1+2+4 instead of single stride 4 (ablation) | AMFR · KITTI-15 train (C+T) | EPE: 3.00 -> 2.86 | Improved |
| use strides 1+2+4 instead of single stride 1 (ablation) | AMFR · KITTI-15 train (C+T) | F1-all: 9.52 -> 9.21 | Improved |
| increase memory length T from 4 to 6 (ablation) | ARFlow · KITTI-15 train (C+T) | EPE: 2.96 -> 2.86 | Improved |
| increase memory length T from 6 to 8 (ablation) | ARFlow · KITTI-15 train (C+T) | EPE: 2.86 -> 2.86 | No clear effect |
| increase GRU iterations K from 4 to 6 (ablation) | ARFlow · KITTI-15 train (C+T) | F1-all: 9.89 -> 9.21 | Improved |
| increase GRU iterations K from 6 to 8 (ablation) | ARFlow · KITTI-15 train (C+T) | F1-all: 9.21 -> 9.23 | No clear effect |
| store memory flows at 1/16 instead of full resolution (ablation) | ARFlow · KITTI-15 train (C+T) | F1-all: 9.27 -> 9.21 | No clear effect |
| replace temporal transformer with pre-trained WAN2.2 (5B) (ablation) | ARFlow · KITTI-15 train (C+T) | F1-all: 9.21 -> 9.56 | Hurt |
| integrate ARFlow into SEA-RAFT(L) | SEA-RAFT(L) · Sintel train Final (C+T) | EPE: 4.11 -> 3.72 | Improved |
Later papers that compared with this method
1 later accepted paper reports ARFlow as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Triangular Consistency as a Universal Constraint for Learning Optical Flow (ECCV 2026 · better on 5 of 6 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- MEMFOF — MEMFOF: High-Resolution Training for Memory-Efficient Multi-Frame Optical Flow Estimation (ICCV 2025) · better on 11 of 12 shared results
- MemFlow — MemFlow: Optical Flow Estimation and Prediction with Memory (CVPR 2024) · better on 3 of 3 shared results
- StreamFlow — StreamFlow: Streamlined Multi-Frame Optical Flow Estimation for Video Sequences (NeurIPS 2024) · better on 3 of 3 shared results
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
- StreamFlow: Streamlined Multi-Frame Optical Flow Estimation for Video Sequences (NeurIPS 2024 · similarity 0.92)
- MemFlow: Optical Flow Estimation and Prediction with Memory (CVPR 2024 · similarity 0.90)
- MESSENGER: Memory-Enhanced Sequential Scene Flow Estimation via Autoregressive Next-Frame Forecasting (NeurIPS 2026 · similarity 0.89)
- DPFlow: Adaptive Optical Flow Estimation with a Dual-Pyramid Framework (CVPR 2025 · similarity 0.89)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.88)
- Taming generative video models for zero-shot optical flow extraction (NeurIPS 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.
