Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-Supervised AutoFlow
Hsin-Ping Huang, Charles Herrmann, Junhwa Hur, Erika Lu, Kyle Sargent, Austin Stone, Ming-Hsuan Yang, Deqing Sun
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 |
|---|---|---|---|
| Use self-supervised losses to search rendering parameters without target-domain flow labels | AutoFlow dataset generation · Sintel and KITTI | AEPE: 7.63 -> 4.22 | Improved |
| Use self-supervised loss-guided rendering instead of supervised AutoFlow | RAFT pre-training · Sintel Clean | AEPE: 1.74 -> 1.83 | Hurt |
| Use self-supervised loss-guided rendering instead of supervised AutoFlow | RAFT pre-training · Sintel Final | AEPE: 2.41 -> 2.59 | Hurt |
| Use self-supervised loss-guided rendering instead of supervised AutoFlow | RAFT pre-training · KITTI | AEPE: 4.18 -> 4.22 | Mixed |
| Add self-supervised fine-tuning after S-AF pre-training | RAFT on Sintel · Sintel Final | AEPE: 2.57 -> 2.40 | Improved |
| Add self-supervised fine-tuning after S-AF pre-training | RAFT on KITTI · KITTI 2015 | Fl-all (%): 13.60 -> 6.56 | Improved |
| Add photometric, smoothness, and distillation terms together to the search metric (ablation) | Self-supervised AutoFlow · Sintel Clean | AEPE: 2.30 -> 1.83 | Improved |
| Add photometric, smoothness, and distillation terms together to the search metric (ablation) | Self-supervised AutoFlow · Sintel Final | AEPE: 2.98 -> 2.59 | Improved |
| Use a single self-supervised loss term instead of combining all three (ablation) | Self-supervised AutoFlow · Sintel Final | AEPE: 2.59 vs 3.24, 2.84, and 3.04 | Hurt |
| Mix the top-3 searched datasets rather than selecting a single candidate (ablation) | Self-supervised AutoFlow · KITTI 2015 | AEPE: 5.40 -> 4.58 | Mixed |
| Use intermediate RAFT predictions in the search metric instead of only the final prediction (ablation) | Self-supervised AutoFlow search · Sintel Final | AEPE | Hurt |
| Use S-AF+SS initialization instead of SMURF for DAVIS segmentation tracking | Optical-flow-based segmentation propagation · DAVIS | IoU: 0.876 -> 0.888 | Improved |
Closest papers by meaning
- SelFlow: Self-Supervised Learning of Optical Flow (CVPR 2019 · similarity 0.88)
- AutoFlow: Learning a Better Training Set for Optical Flow (CVPR 2021 · similarity 0.87)
- Semi-Supervised Learning of Optical Flow by Flow Supervisor (ECCV 2022 · similarity 0.87)
- Learning by Distillation: A Self-Supervised Learning Framework for Optical Flow Estimation (TPAMI 2021 · similarity 0.85)
- No More Shortcuts: Realizing the Potential of Temporal Self-Supervision (AAAI 2024 · similarity 0.83)
- ZeroFlow: Scalable Scene Flow via Distillation (ICLR 2024 · similarity 0.83)
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.
