Research map / Video / Video segmentation and tracking
Unsupervised Learning of Multi-Frame Optical Flow with Occlusions
Joel Janai, Fatma Guney, Anurag Ranjan, Michael Black, Andreas Geiger
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 |
|---|---|---|---|
| extend two-frame PWC-Net to three frames (ablation) | PWC-Net · RoamingImages test | EPE (all): 14.14 -> 10.11 | Improved |
| add occlusion reasoning to the multi-frame model (ablation) | Multi · RoamingImages test | EPE (all): 10.11 -> 6.93 | Improved |
| add occlusion reasoning to the multi-frame model (ablation) | Multi · RoamingImages test | EPE (occluded): 18.22 -> 8.55 | Improved |
| add hard constant velocity constraint (ablation) | Ours-None · RoamingImages test | EPE (occluded): 16.26 -> 8.55 | Improved |
| add hard constant velocity constraint (ablation) | Ours-None · RoamingImages test | EPE (all): 8.37 -> 6.93 | Improved |
| add soft constant velocity constraint (ablation) | Ours-None · RoamingImages test | EPE (all): 8.37 -> 8.17 | No clear effect |
| fine-tune with hard constraint on MPI Sintel (ablation) | Ours-Hard · MPI Sintel Clean train | EPE (all): 5.38 -> 6.05 | Hurt |
| use soft instead of no temporal constraint in fine-tuning (ablation) | Ours-None-ft · MPI Sintel Clean train | EPE (all): 4.74 -> 3.89 | Improved |
| use soft instead of no temporal constraint in fine-tuning (ablation) | Ours-None-ft · KITTI 2015 train | EPE (all): 6.65 -> 6.59 | No clear effect |
| use soft instead of no temporal constraint in fine-tuning (ablation) | Ours-None-ft · MPI Sintel Final train | EPE (occluded): 12.66 -> 12.87 | Mixed |
| pretrain UnFlow-CSS on RoamingImages instead of its default data (control for dataset) | UnFlow-CSS · KITTI 2015 train | EPE (all): 8.10 -> 19.26 | Hurt |
| use the multi-frame model pretrained only (no fine-tuning) vs prior pretrained best | Ours-Hard · MPI Sintel Final train | EPE (all): 6.34 -> 6.01 | Improved |
Closest papers by meaning
- Occlusion Aware Unsupervised Learning of Optical Flow (CVPR 2018 · similarity 0.92)
- What Matters in Unsupervised Optical Flow (ECCV 2020 · similarity 0.91)
- Learning by Analogy: Reliable Supervision From Transformations for Unsupervised Optical Flow Estimation (CVPR 2020 · similarity 0.91)
- Learning To Estimate Hidden Motions With Global Motion Aggregation (ICCV 2021 · similarity 0.90)
- OCAI: Improving Optical Flow Estimation by Occlusion and Consistency Aware Interpolation (CVPR 2024 · similarity 0.89)
- UnSAMFlow: Unsupervised Optical Flow Guided by Segment Anything Model (CVPR 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.
