Research map / 3D vision / Depth and stereo
Uncertainty Estimates and Multi-Hypotheses Networks for Optical Flow
Eddy Ilg, Ozgun Cicek, Silvio Galesso, Aaron Klein, Osama Makansi, Frank Hutter, Thomas Brox
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 with predictive negative log-likelihood loss instead of EPE loss (Dropout) (ablation) | Dropout ensemble · Sintel train clean | AUSE: 0.212 -> 0.158 | Improved |
| train with predictive loss instead of EPE loss (Bootstrapped Ensemble) (ablation) | BootstrappedEnsemble · Sintel train clean | AUSE: 0.209 -> 0.127 | Improved |
| train with predictive loss instead of EPE loss (SGDR Ensemble) (ablation) | SGDREnsemble · Sintel train clean | AUSE: 0.191 -> 0.134 | Improved |
| use predictive instead of empirical training (ablation) | Ensembles · Sintel train clean | EPE: 3.41 -> 3.46 | Hurt |
| add a merging network on top of Bootstrapped Ensemble (ablation) | BootstrappedEnsemble (Pred) · Sintel train clean | AUSE: 0.127 -> 0.102 | Improved |
| add a merging network on top of Bootstrapped Ensemble (ablation) | BootstrappedEnsemble (Pred) · Sintel train clean | EPE: 3.46 -> 3.20 | Improved |
| replace Bootstrapped Ensemble with multi-hypothesis FlowNetH plus merging network | BootstrappedEnsemble-Merged · Sintel train clean | runtime: 332ms -> 60ms | Improved |
| replace Bootstrapped Ensemble with FlowNetH plus merging network | BootstrappedEnsemble-Merged · Sintel train clean | AUSE: 0.102 -> 0.095 | No clear effect |
| use ensemble of 8 instead of single predictive network (ablation) | FlowNetC (Pred) · Sintel train clean | AUSE: 0.133 -> 0.134 | No clear effect |
| use Dropout instead of other ensemble types | predictive ensembles · Sintel train clean | AUSE: 0.158 vs 0.127 | Hurt |
| stack one refinement network (S) on FlowNetH-Pred-Merged (ablation) | FlowNetH-Pred-Merged · Sintel train clean | AUSE: 0.117 -> 0.091 | Improved |
| stack two refinement networks (SS) on FlowNetH-Pred-Merged (ablation) | FlowNetH-Pred-Merged · KITTI 2012+2015 validation split | EPE: 7.84 -> 6.50 | Improved |
Closest papers by meaning
- U^2Flow: Uncertainty-Aware Unsupervised Optical Flow Estimation (CVPR 2026 · similarity 0.90)
- ProbFlow: Joint Optical Flow and Uncertainty Estimation (ICCV 2017 · similarity 0.90)
- Learning Optical Flow From Still Images (CVPR 2021 · similarity 0.89)
- Models Matter, So Does Training: An Empirical Study of CNNs for Optical Flow Estimation (TPAMI 2019 · similarity 0.89)
- Context-Aware Iteration Policy Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.88)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.87)
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.
