Research map / Label-efficient and robust learning / Self-supervised and contrastive
Conditional Prior Networks for Optical Flow
Yanchao Yang, Stefano Soatto
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 |
|---|---|---|---|
| train FlowNetS with learned conditional prior, unsupervised, on Flying Chairs (CPNFlow-C vs FlowNetS) | FlowNetS · KITTI 2012 Train | EPE: 8.26 -> 7.33 | Improved |
| train with conditional prior on Flying Chairs instead of supervised FlowNetC | FlowNetC · KITTI 2012 Train | EPE: 9.35 -> 7.33 | Improved |
| train on KITTI with the conditional prior (CPNFlow-K) vs supervised FlowNet2 | FlowNet2 · KITTI 2012 Train | EPE: 4.09 -> 3.76 | Improved |
| train on real KITTI videos (CPNFlow-K) instead of Flying Chairs (CPNFlow-C) | CPNFlow · KITTI 2012 Train | EPE: 7.33 -> 3.76 | Improved |
| train on KITTI instead of Flying Chairs | CPNFlow · Sintel Train clean | EPE: 4.87 -> 6.46 | Hurt |
| add occlusion reasoning to CPNFlow-K (ablation) | CPNFlow-K · KITTI 2012 Train | EPE: 3.76 -> 3.11 | Improved |
| add occlusion reasoning to CPNFlow-K (ablation) | CPNFlow-K · KITTI 2015 Train | EPE: 9.63 -> 7.82 | Improved |
| add occlusion reasoning to CPNFlow-K (ablation) | CPNFlow-K · Sintel Train final | EPE: 7.12 -> 7.52 | Hurt |
| use learned conditional prior (CPNFlow-K) instead of UnFlowC unsupervised loss | UnFlowC · KITTI 2012 Train | EPE: 7.11 -> 3.76 | Improved |
| use learned prior with occlusion (CPNFlow-K-o) vs UnFlowC-oc | UnFlowC-oc · KITTI 2012 Train | EPE: 3.78 -> 3.11 | Improved |
| use learned prior (CPNFlow-C) instead of LDOF variational regularizer | LDOF · KITTI 2012 Train | EPE: 13.7 -> 7.33 | Improved |
| use learned prior CPNFlow-C vs supervised FlowNetS on Sintel clean | FlowNetS · Sintel Train clean | EPE: 4.50 -> 4.87 | No clear effect |
Closest papers by meaning
- Uncertainty Estimates and Multi-Hypotheses Networks for Optical Flow (ECCV 2018 · similarity 0.89)
- Neural Scene Flow Prior (NeurIPS 2021 · similarity 0.88)
- Learning Optical Flow From Still Images (CVPR 2021 · similarity 0.87)
- A Lightweight Optical Flow CNN —Revisiting Data Fidelity and Regularization (TPAMI 2020 · similarity 0.87)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.87)
- Context-Aware Iteration Policy 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.
