Research map / Detection and segmentation / Open-set and few-shot detection
Self-Supervised Human Detection and Segmentation via Background Inpainting
Isinsu Katircioglu, Helge Rhodin, Victor Constantin, Jörg Spörri, Mathieu Salzmann, Pascal Fua
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 the L_v mask prior (vs none) (ablation) | Ours w/o optical flow · Ski PTZ dataset (validation) | J measure: 0.51 -> 0.67 | Improved |
| use L_v prior instead of L1 prior (ablation) | Ours w/o optical flow · Ski PTZ dataset (validation) | J measure: 0.62 -> 0.67 | Improved |
| use L_v prior instead of L2 prior (ablation) | Ours w/o optical flow · Ski PTZ dataset (validation) | J measure: 0.61 -> 0.67 | Improved |
| add ImageNet pre-training (vs random init) (ablation) | Ours w/o optical flow · Ski PTZ dataset (validation) | J measure: 0.60 -> 0.67 | Improved |
| ImageNet weights vs unsupervised pre-training (ablation) | Ours w/o optical flow · Ski PTZ dataset (validation) | J measure: 0.62 -> 0.67 | No clear effect |
| add optical flow supervision during training (ablation) | Ours · Ski PTZ dataset (test) | J measure: 0.62 -> 0.70 | Improved |
| add optical flow supervision during training (ablation) | Ours · Handheld dataset (test) | J measure: 0.75 -> 0.70 | Hurt |
| add optical flow supervision during training (ablation) | Ours · Iceskating dataset (test) | J measure: 0.66 -> 0.69 | Improved |
| add CRF post-processing (ablation) | Ours w/ optical flow · Ski PTZ dataset (test) | F measure: 0.77 -> 0.83 | Improved |
| add CRF post-processing (ablation) | Ours w/ optical flow · Handheld dataset (test) | J measure: 0.70 -> 0.76 | Improved |
| use Monte Carlo importance sampling instead of Gumbel-Softmax (ablation) | Ours · Human3.6M (detection) | mAP@0.5: 0.30 -> 0.58 | Improved |
| use importance sampling instead of uniform sampling (ablation) | Ours · Human3.6M (detection) | convergence | Improved |
Closest papers by meaning
- Learning Accurate Segmentation Purely from Self-Supervision (ECCV 2026 · similarity 0.90)
- Self-Supervised Learning of Object Parts for Semantic Segmentation (CVPR 2022 · similarity 0.89)
- A Survey of Self-Supervised and Few-Shot Object Detection (TPAMI 2022 · similarity 0.89)
- UniT: Unified Knowledge Transfer for Any-Shot Object Detection and Segmentation (CVPR 2021 · similarity 0.89)
- Object Discovery and Representation Networks (ECCV 2022 · similarity 0.89)
- Human Detection and Segmentation via Multi-View Consensus (ICCV 2021 · 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.
