Research map / Probabilistic and scientific modeling / Molecules and proteins
Improving Optical Flow on a Pyramid Level
Markus Hofinger, Samuel Rota Bulò, Lorenzo Porzi, Arno Knapitsch, Thomas Pock, Peter Kontschieder
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 gradient stopping between pyramid levels (ablation) | HD3 re-trained · Flying Chairs2 val | EPE: 1.422 -> 1.215 | Improved |
| add flow cues on top of gradient stopping and sampling (ablation) | HD3 with gradient stopping + sampling · Flying Chairs2 val | EPE: 1.208 -> 1.186 | Improved |
| add flow cues on top of gradient stopping with warping (ablation) | HD3 with gradient stopping, warping · Flying Chairs2 val | EPE: 1.215 -> 1.216 | No clear effect |
| add loss max-pooling to the full model (ablation) | HD3 with all other contributions · Flying Chairs2 val | Fl-all (%): 6.15 -> 5.97 | Improved |
| add SAD distance on top of sampling and flow cues (ablation) | HD3 with gradient stopping, sampling, cues · Flying Things val (trained on Flying Chairs2) | EPE: 19.616 -> 15.136 | Improved |
| add LMP to gradient stopping alone (no sampling, cues or SAD) (ablation) | HD3 with gradient stopping · Sintel final train (trained on Flying Chairs2) | EPE: 5.774 -> 12.529 | Hurt |
| add cost volume sampling on top of gradient stopping (ablation) | HD3 with gradient stopping · Flying Things val (trained on Flying Chairs2) | Fl-all (%): 26.84 -> 24.75 | Improved |
| extend cost volume range from +-4 to +-8 with sampling and LMP (ablation) | HD3 sampling + LMP · Flying Chairs2 val | EPE: 0.06 | Improved |
| extend cost volume range from +-4 to +-8 with sampling, no LMP (ablation) | HD3 sampling · Flying Chairs2 val | EPE: 0.01 | No clear effect |
| add knowledge distillation to KITTI fine-tuning (ablation) | IOFPL-CVr8 · KITTI 2015 train | Fl-all (%): 3.58 -> 3.28 | Improved |
| add knowledge distillation to KITTI fine-tuning (generalization) (ablation) | IOFPL-CVr8 · Flying Things val | Fl-all (%): 46.2 -> 38.0 | Improved |
| add gradient stopping at the upsampling layer (ablation) | PWC-Net (IRR-PWC code) on Flying Chairs · Flying Chairs val | EPE: 10% at 20 epochs, 6% at 100 epochs | Improved |
Closest papers by meaning
- Optical Flow Estimation Using a Spatial Pyramid Network (CVPR 2017 · similarity 0.93)
- DPFlow: Adaptive Optical Flow Estimation with a Dual-Pyramid Framework (CVPR 2025 · similarity 0.89)
- A Lightweight Optical Flow CNN —Revisiting Data Fidelity and Regularization (TPAMI 2020 · similarity 0.89)
- SKFlow: Learning Optical Flow with Super Kernels (NeurIPS 2022 · similarity 0.89)
- DDFlow: Learning Optical Flow with Unlabeled Data Distillation (AAAI 2019 · similarity 0.89)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · 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.
