Research map / 3D vision / Depth and stereo
What Matters in Unsupervised Optical Flow
Rico Jonschkowski, Austin Stone, Jonathan T. Barron, Ariel Gordon, Kurt Konolige, Anelia Angelova
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 self-supervision (ablation) | UFlow · KITTI-15 train (all EPE) | EPE: 4.15 -> 2.94 | Improved |
| add occlusion masking (ablation) | UFlow · KITTI-15 train (all EPE) | EPE: 3.28 -> 2.94 | Improved |
| add stop-gradient at range-map occlusion mask (ablation) | UFlow · all datasets | divergence: div. -> 3.10 | Improved |
| add cost volume normalization and level dropout (ablation) | UFlow (full) · KITTI-15 train (all EPE) | EPE: 39.09 -> 2.94 | Improved |
| add cost volume normalization and level dropout (ablation) | UFlow (minimal) · KITTI-15 train (all EPE) | EPE: 13.30 -> 13.07 | No clear effect |
| use Census instead of Charbonnier photometric loss (ablation) | UFlow (full) · KITTI-15 train (all EPE) | EPE: 5.56 -> 3.08 | Improved |
| use SSIM instead of Census loss (ablation) | UFlow (minimal) · Sintel Final train | EPE: 6.98 -> 5.41 | Improved |
| apply smoothness at flow resolution (level 2) instead of level 0 (ablation) | UFlow (full) · Flying Chairs test | EPE: 2.87 -> 2.58 | Improved |
| add image resizing to self-supervision crops (ablation) | UFlow (full) · KITTI-15 train (all EPE) | EPE: 3.53 -> 2.86 | Improved |
| use shared teacher instead of frozen teacher (ablation) | UFlow (full) · KITTI-15 train (all EPE) | EPE: 2.99 -> 2.86 | Improved |
| use 640x640 square input instead of 384x1280 (ablation) | UFlow (full) · KITTI-15 train (all EPE) | EPE: 3.80 -> 2.93 | Improved |
| add pretraining on Chairs (ablation) | UFlow (full) · Sintel Clean train | EPE: 3.38 -> 3.00 | Improved |
Closest papers by meaning
- UnSAMFlow: Unsupervised Optical Flow Guided by Segment Anything Model (CVPR 2024 · similarity 0.91)
- UnFlow: Unsupervised Learning of Optical Flow With a Bidirectional Census Loss (AAAI 2018 · similarity 0.91)
- Unsupervised Learning of Multi-Frame Optical Flow with Occlusions (ECCV 2018 · similarity 0.91)
- Learning by Analogy: Reliable Supervision From Transformations for Unsupervised Optical Flow Estimation (CVPR 2020 · similarity 0.91)
- Occlusion Aware Unsupervised Learning of Optical Flow (CVPR 2018 · similarity 0.90)
- U^2Flow: Uncertainty-Aware Unsupervised Optical Flow Estimation (CVPR 2026 · similarity 0.90)
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.
