Research map / 3D vision / Depth and stereo
ProbFlow: Joint Optical Flow and Uncertainty Estimation
Anne S. Wannenwetsch, Margret Keuper, Stefan Roth
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 |
|---|---|---|---|
| replace best post-hoc measure (Learned) with ProbFlow marginal entropy | uncertainty estimate · Middlebury training | AUC: 0.496 -> 0.466 | Improved |
| replace Learned uncertainty with ProbFlow marginal entropy | uncertainty estimate · Middlebury training | Spearman CC: 0.324 -> 0.374 | Improved |
| replace Energy/Learned uncertainty with ProbFlowFields entropy | uncertainty estimate · Sintel validation | AUC: 0.474 -> 0.398 | Improved |
| replace Learned uncertainty with ProbFlowFields entropy | uncertainty estimate · Sintel validation | Spearman CC: 0.451 -> 0.563 | Improved |
| replace MAP FlowFields* with probabilistic inference (ProbFlowFields) | FlowFields energy · Sintel validation | AEPE: 3.161 -> 3.127 | Improved |
| use probabilistic ClassicA instead of ClassicA | ClassicA energy · Middlebury training | AEPE: 0.295 -> 0.296 | No clear effect |
| apply bilateral solver guided by uncertainties on ProbFlowFields | ProbFlowFields · Sintel validation | AEPE: 3.127 -> 3.052 | Improved |
| add uncertainty-based track termination (3) to FB check (1) (ablation) | trajectory generation · FBMS-59 training | F-measure: 78.03% -> 80.92% | Improved |
| use ProbFlowFields instead of FlowFields in trajectory generation | motion segmentation · FBMS-59 training | point density: 0.83% -> 1.18% | Improved |
| jointly estimate uncertainty (ProbFlowFields) instead of FlowFields plus learned post-hoc measure | runtime · Sintel validation | runtime (s): 123.8s -> 38.1s | Improved |
| add uncertainty estimation to FlowFields | runtime · Sintel validation | runtime (s): 19.9s -> 38.1s | Hurt |
| use ProbFlowFields dense segmentation instead of FlowFields dense | motion segmentation · FBMS-59 test (dense) | F-measure: 70.83% -> 75.05% | Improved |
Closest papers by meaning
- U^2Flow: Uncertainty-Aware Unsupervised Optical Flow Estimation (CVPR 2026 · similarity 0.91)
- Uncertainty Estimates and Multi-Hypotheses Networks for Optical Flow (ECCV 2018 · similarity 0.90)
- FlowPainter: Inpainting Optical Flow via Confidence-Guided Completion (ECCV 2026 · similarity 0.89)
- DifFlow3D: Toward Robust Uncertainty-Aware Scene Flow Estimation with Iterative Diffusion-Based Refinement (CVPR 2024 · similarity 0.87)
- Shape and Texture: What Influences Reliable Optical Flow Estimation? (CVPR 2025 · similarity 0.86)
- DPFlow: Adaptive Optical Flow Estimation with a Dual-Pyramid Framework (CVPR 2025 · 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.
