Research map / 3D vision / Depth and stereo
UnFlow: Unsupervised Learning of Optical Flow With a Bidirectional Census Loss
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 brightness constancy with the census transform (ablation) | UnFlow-C unsupervised loss · KITTI 2012 train | AEE (All): 7.20 -> 4.66 | Helped |
| Replace first-order with second-order smoothness (ablation) | UnFlow-C unsupervised loss · KITTI 2012 train | AEE (All): 7.20 -> 7.11 | Helped |
| Add occlusion masking to the census and second-order smoothness loss (ablation) | UnFlow-C unsupervised loss · KITTI 2012 train | AEE (All): 4.40 -> 4.32 | Helped |
| Add forward-backward consistency to census loss, second-order smoothness, and occlusion masking (ablation) | UnFlow-C unsupervised loss · KITTI 2012 train | AEE (All): 4.32 -> 3.78 | Helped |
| Use a second-order rather than first-order smoothness penalty (ablation) | UnFlow-C unsupervised loss · KITTI 2015 train | Fl-all: 41.43% -> 38.64% | Helped |
| Train with the full bidirectional census loss instead of the brightness-constancy baseline (ablation) | UnFlow-C · KITTI 2012 train | AEE (All): 7.20 -> 3.78 | Helped |
| Add a FlowNetS refinement network, forming UnFlow-CS (ablation) | UnFlow-C · KITTI 2012 train | AEE (All): 3.78 -> 3.30 | Helped |
| Add a second FlowNetS refinement network, forming UnFlow-CSS (ablation) | UnFlow-CS · KITTI 2012 train | AEE (All): 3.30 -> 3.29 | Helped |
| Fine-tune the unsupervised stack on sparse KITTI ground truth (ablation) | UnFlow-CSS · KITTI 2012 test | AEE (All): 3.29 -> 1.7 | Helped |
| Train on Cityscapes rather than KITTI imagery (ablation) | UnFlow-C · KITTI 2012 train | AEE (All): 3.78 -> 5.08 | Hurt |
Closest papers by meaning
- Unsupervised Deep Learning for Optical Flow Estimation (AAAI 2017 · similarity 0.92)
- Learning by Analogy: Reliable Supervision From Transformations for Unsupervised Optical Flow Estimation (CVPR 2020 · similarity 0.92)
- UnSAMFlow: Unsupervised Optical Flow Guided by Segment Anything Model (CVPR 2024 · similarity 0.91)
- DDFlow: Learning Optical Flow with Unlabeled Data Distillation (AAAI 2019 · similarity 0.91)
- OCAI: Improving Optical Flow Estimation by Occlusion and Consistency Aware Interpolation (CVPR 2024 · similarity 0.88)
- Taming generative video models for zero-shot optical flow extraction (NeurIPS 2025 · 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.