Research map / Architectures and efficiency / Neural architecture search
FlowNAS: Neural Architecture Search for Optical Flow Estimation
Zhiwei Lin, Tingting Liang, Taihong Xiao, Yongtao Wang, Ming–Hsuan Yang
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 |
|---|---|---|---|
| use SepConv search space instead of Conv (ablation) | FlowNAS · KITTI-15 trainval (C+T) | F1-all (%): 17.4 -> 16.9 | Improved |
| use SepConv search space instead of ShuffleConv (ablation) | FlowNAS · KITTI-15 trainval (C+T) | F1-all (%): 17.9 -> 16.9 | Improved |
| use Channel Maximize instead of Dynamic Channel Projection for feature alignment (ablation) | FlowNAS FAD · KITTI-15 val | F1-all (%): 4.80 -> 3.63 | Improved |
| use Channel Maximize instead of Spatial Attention (ablation) | FlowNAS FAD · KITTI-15 val | F1-all (%): 83.0 -> 3.63 | Improved |
| use Channel Maximize instead of Channel Average (ablation) | FlowNAS FAD · KITTI-15 val | F1-all (%): 3.89 -> 3.63 | Improved |
| add feature alignment distillation (FAD) to super-network training (ablation) | FlowNAS-RAFT · KITTI-15 val | F1-all (%): 4.05 -> 3.63 | Improved |
| add FAD when training sub-network from scratch (ablation) | FlowNAS-RAFT · KITTI-15 val | F1-all (%): 3.84 -> 3.78 | Improved |
| add rough decoder supervision L_rough (ablation) | FlowNAS super-network · Sintel val Final | EPE: 3.00 -> 3.91 | Hurt |
| replace flow loss with rough decoder loss only (ablation) | FlowNAS super-network · KITTI-15 val | F1-all (%): 3.63 -> 98.7 | Hurt |
| add FAD loss L_D to flow loss (ablation) | FlowNAS super-network · KITTI-15 val | F1-all (%): 4.05 -> 3.63 | Improved |
| apply FlowNAS encoder search to GMA | GMA · KITTI-15 trainval (C+T) | F1-all (%): 17.1 -> 16.9 | Improved |
| apply FlowNAS-RAFT-S encoder to PWC-Net | PWC-Net · Sintel trainval (C+T) | EPE Clean: 2.55 -> 1.83 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.
- RAFT — RAFT: Recurrent All-Pairs Field Transforms for Optical Flow (ECCV 2020) · better on 6 of 6 shared results
- SeparableFlow — Separable Flow: Learning Motion Cost Volumes for Optical Flow Estimation (ICCV 2021) · better on 0 of 4 shared results
- PWC-Net — PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume (CVPR 2018) · 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
- LiteFlowNet: A Lightweight Convolutional Neural Network for Optical Flow Estimation (CVPR 2018 · similarity 0.90)
- FlowNet 2.0: Evolution of Optical Flow Estimation With Deep Networks (CVPR 2017 · similarity 0.89)
- Models Matter, So Does Training: An Empirical Study of CNNs for Optical Flow Estimation (TPAMI 2019 · similarity 0.89)
- A Lightweight Optical Flow CNN —Revisiting Data Fidelity and Regularization (TPAMI 2020 · similarity 0.88)
- Context-Aware Iteration Policy Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.86)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · 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.
