Research map / Architectures and efficiency / CNN design
PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume
Deqing Sun, Xiaodong Yang, Ming-Yu Liu, Jan Kautz
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 PWC-Net instead of FlowNet2.0 | Optical flow estimation · MPI Sintel final | EPE: 3.93 vs. 6.02 | Improved |
| Fine-tune PWC-Net on KITTI training data | PWC-Net · KITTI 2015 test | Fl-all: 9.60% | Improved |
| Add a context network (ablation) | PWC-Net · MPI Sintel final | EPE: 4.74 -> 4.59 | Improved |
| Add DenseNet connections after fine-tuning on FlyingThings3D (ablation) | PWC-Net · MPI Sintel final | EPE: 4.08 -> 3.95 | Improved |
| Add warping layers (ablation) | PWC-Net · MPI Sintel final | EPE: 5.30 -> 4.59 | Improved |
| Use a residual optical-flow estimator (ablation) | PWC-Net · MPI Sintel final | EPE: 4.59 -> 4.43 | Improved |
| Increase feature pyramid extractor capacity (ablation) | PWC-Net · MPI Sintel final | EPE: 4.59 -> 4.17 | Improved |
| Decrease feature pyramid extractor capacity (ablation) | PWC-Net · MPI Sintel final | EPE: 4.59 -> 4.56 | Hurt |
| Increase optical-flow estimator capacity (ablation) | PWC-Net · MPI Sintel final | EPE: 4.59 -> 4.50 | Improved |
| Reduce cost-volume search range to 2 (ablation) | PWC-Net · MPI Sintel final | EPE: 4.59 -> 4.50 | Improved |
| Increase cost-volume search range to 6 (ablation) | PWC-Net · MPI Sintel final | EPE: 4.59 -> 4.60 | No clear effect |
Later papers that compared with this method
9 later accepted papers report PWC-Net as a baseline in their results tables; 9 of them report a better number on most of the results they share with it.
- Removing Cost Volumes from Optical Flow Estimators (ICCV 2025 · better on 1 of 1 shared result)
- FlowNAS: Neural Architecture Search for Optical Flow Estimation (IJCV 2023 · better on 1 of 1 shared result)
- Disentangling Architecture and Training for Optical Flow (ECCV 2022 · better on 3 of 3 shared results)
- Separable Flow: Learning Motion Cost Volumes for Optical Flow Estimation (ICCV 2021 · better on 1 of 1 shared result)
- A Lightweight Optical Flow CNN —Revisiting Data Fidelity and Regularization (TPAMI 2020 · better on 4 of 4 shared results)
- GLU-Net: Global-Local Universal Network for Dense Flow and Correspondences (CVPR 2020 · better on 1 of 1 shared result)
- LSM: Learning Subspace Minimization for Low-Level Vision (CVPR 2020 · better on 1 of 1 shared result)
- ScopeFlow: Dynamic Scene Scoping for Optical Flow (CVPR 2020 · better on 1 of 1 shared result)
1 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
- Models Matter, So Does Training: An Empirical Study of CNNs for Optical Flow Estimation (TPAMI 2019 · similarity 0.92)
- Disentangling Architecture and Training for Optical Flow (ECCV 2022 · similarity 0.89)
- Optical Flow Estimation Using a Spatial Pyramid Network (CVPR 2017 · similarity 0.88)
- A Lightweight Optical Flow CNN —Revisiting Data Fidelity and Regularization (TPAMI 2020 · similarity 0.87)
- LiteFlowNet: A Lightweight Convolutional Neural Network for Optical Flow Estimation (CVPR 2018 · similarity 0.86)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.84)
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.
