Research map / Architectures and efficiency / RNNs and normalization
Models Matter, So Does Training: An Empirical Study of CNNs for Optical Flow Estimation
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 learned feature pyramid instead of image pyramid (ablation) | PWC-Net · MPI Sintel Final train (trained on FlyingChairs) | AEPE: 5.58 -> 4.59 | Improved |
| increase feature extractor to three layers per level (ablation) | PWC-Net · MPI Sintel Final train (trained on FlyingChairs) | AEPE: 4.59 -> 4.17 | Improved |
| add warping layer (ablation) | PWC-Net · MPI Sintel Final train (trained on FlyingChairs) | AEPE: 5.30 -> 4.59 | Improved |
| add context network (ablation) | PWC-Net · MPI Sintel Final train (trained on FlyingChairs) | AEPE: 4.74 -> 4.59 | Improved |
| add DenseNet connections (ablation) | PWC-Net · MPI Sintel Final train (trained on FlyingChairs) | AEPE: 4.37 -> 4.59 | Mixed |
| add cost volume (search range 4 vs 0) (ablation) | PWC-Net · MPI Sintel Final train (trained on FlyingChairs) | AEPE: 5.09 -> 4.59 | Improved |
| deepen optical flow estimator to seven layers (ablation) | PWC-Net · MPI Sintel Final train (trained on FlyingChairs) | AEPE: 4.59 -> 4.50 | Improved |
| make flow estimator residual (ablation) | PWC-Net · MPI Sintel Final train (trained on FlyingChairs) | AEPE: 4.59 -> 4.43 | Improved |
| use 6-level instead of 7-level pyramid (ablation) | PWC-Net · MPI Sintel Final train (trained on FlyingChairs) | AEPE: 4.59 -> 4.32 | Mixed |
| add FlyingThings3D fine-tuning after FlyingChairs (ablation) | PWC-Net · MPI Sintel Final train | AEPE: 4.59 -> 3.93 | Improved |
| retrain FlowNetC with PWC-Net training procedure (FlowNetC+) vs published FlowNet2 | FlowNetC · MPI Sintel Final test | AEPE: 6.02 -> 5.47 | Improved |
| fix KITTI I/O bug and retrain on all 200 sequences | PWC-Net-ft · KITTI 2015 test | Fl-all (%): 9.60% -> 7.90% | Improved |
Closest papers by meaning
- LiteFlowNet: A Lightweight Convolutional Neural Network for Optical Flow Estimation (CVPR 2018 · similarity 0.93)
- A Lightweight Optical Flow CNN —Revisiting Data Fidelity and Regularization (TPAMI 2020 · similarity 0.93)
- PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume (CVPR 2018 · similarity 0.92)
- FlowNet 2.0: Evolution of Optical Flow Estimation With Deep Networks (CVPR 2017 · similarity 0.92)
- Optical Flow Estimation Using a Spatial Pyramid Network (CVPR 2017 · similarity 0.90)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.89)
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.
