Research map / Label-efficient and robust learning / Self-supervised and contrastive
Complementing Brightness Constancy with Deep Networks for Optical Flow Prediction
Vincent Le Guen, Clément Rambour, Nicolas Thome
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 the BC/augmentation decomposition in place of RAFT | RAFT · FlyingChairs | EPE: 0.818 -> 0.743 | Improved |
| add a second RAFT branch mixed equally (ensembling/more parameters) (ablation) | RAFT · FlyingChairs | EPE: 0.818 -> 0.800 | Improved |
| add naive BC photometric loss to a single RAFT branch (ablation) | RAFT · FlyingChairs | EPE: 0.818 -> 0.794 | Improved |
| add uncertainty weighting alpha to the linear decomposition (ablation) | COMBO · FlyingChairs | EPE: 0.769 -> 0.743 | Improved |
| add minimization of norm of (w_a,w_p) for unique decomposition (ablation) | COMBO · FlyingChairs | EPE: 0.752 -> 0.743 | Improved |
| replace RAFT with COMBO w_p on BC-certain zones | RAFT · FlyingChairs (BC certain zones) | weighted EPE: 0.48 -> 0.30 | Improved |
| replace RAFT with COMBO w_a on BC-uncertain zones | RAFT · FlyingChairs (BC uncertain zones) | weighted EPE: 2.29 -> 0.52 | Improved |
| add semi-supervised training with unlabelled KITTI frames | COMBO · KITTI test-resplit (Chairs checkpoint) | Fl-epe: 2.19 -> 1.75 | Improved |
| add semi-supervised training with unlabelled frames from Sintel checkpoint | COMBO · KITTI test-resplit (Sintel checkpoint) | Fl-epe: 1.75 -> 1.74 | No clear effect |
| reduce labelled KITTI training images to 10 percent, COMBO-semisup vs RAFT | COMBO · KITTI-resplit (from Chairs) | EPE: 11.6 -> 4.4 | Improved |
| replace RAFT with COMBO at Chairs stage | RAFT · Sintel-test-resplit Final (Chairs stage) | EPE: 1.76 -> 1.58 | Improved |
| replace RAFT with COMBO at Chairs+Things stage | RAFT · Sintel-test-resplit Final (Chairs+Things stage) | EPE: 1.17 -> 0.96 | Improved |
Closest papers by meaning
- Uncertainty Estimates and Multi-Hypotheses Networks for Optical Flow (ECCV 2018 · similarity 0.88)
- Unsupervised Deep Learning for Optical Flow Estimation (AAAI 2017 · similarity 0.88)
- Models Matter, So Does Training: An Empirical Study of CNNs for Optical Flow Estimation (TPAMI 2019 · similarity 0.88)
- DDFlow: Learning Optical Flow with Unlabeled Data Distillation (AAAI 2019 · similarity 0.87)
- Conditional Prior Networks for Optical Flow (ECCV 2018 · similarity 0.87)
- 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.
