Research map / Label-efficient and robust learning / Domain adaptation and generalization
Leveraging Self-Supervision for Cross-Domain Crowd Counting
Check 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 |
|---|---|---|---|
| Apply the two-stage self-supervision approach | SFCN cross-domain crowd counting · ShanghaiTech Part A | MAE: 121 -> 109.2 | Helped |
| Apply the two-stage self-supervision approach | SFCN cross-domain crowd counting · ShanghaiTech Part B | MAE: 12.8 -> 11.4 | Helped |
| Apply the two-stage self-supervision approach | SFCN cross-domain crowd counting · UCF_CC_50 | MAE: 355 -> 336.5 | Helped |
| Apply the two-stage self-supervision approach | SFCN cross-domain crowd counting · UCF-QNRF | MAE: 210 -> 198.3 | Helped |
| Apply the two-stage self-supervision approach | SFCN cross-domain crowd counting · WorldExpo'10 | MAE (average): 20.4 -> 16.6 | Helped |
| Combine image-wise pretraining with pixel-wise pseudo-label training rather than use either stage alone (ablation) | SFCN on UCF-QNRF · UCF-QNRF | MAE: 208.3 -> 198.3 | Helped |
| Add image-wise self-supervision to pixel-wise self-supervision (ablation) | SFCN on UCF-QNRF · UCF-QNRF | MAE: 208.3 -> 198.3 | Helped |
| Also flip synthetic images for image-wise supervision (ablation) | OURS-IMG on UCF-QNRF · UCF-QNRF | MAE: 242.8 -> 243.0 | No clear effect |
| Keep image-wise supervision during the second training stage (ablation) | OURS on UCF-QNRF · UCF-QNRF | MAE: 198.3 -> 198.5 | No clear effect |
| Discard the most uncertain 10% of pseudo labels (α = 10) rather than retain all (ablation) | SFCN self-training on UCF-QNRF · UCF-QNRF | MAE: 229.3 -> 198.3 | Helped |
| Add Masksembles without domain adaptation (ablation) | SFCN on UCF-QNRF · UCF-QNRF | MAE: 275.5 -> 273.1 | No clear effect |
| Use upside-down flipping rather than no image-wise supervision (ablation) | SFCN on UCF-QNRF · UCF-QNRF | MAE: 208.3 -> 198.3 | Helped |
Closest papers by meaning
- Learning From Synthetic Data for Crowd Counting in the Wild (CVPR 2019 · similarity 0.91)
- Context-Aware Crowd Counting (CVPR 2019 · similarity 0.91)
- Almost Unsupervised Learning for Dense Crowd Counting (AAAI 2019 · similarity 0.90)
- Revisiting Perspective Information for Efficient Crowd Counting (CVPR 2019 · similarity 0.90)
- Divide and Grow: Capturing Huge Diversity in Crowd Images With Incrementally Growing CNN (CVPR 2018 · similarity 0.90)
- Single Domain Generalization for Crowd Counting (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.