Research map / Label-efficient and robust learning / Domain adaptation and generalization
Bi-Level Alignment for Cross-Domain Crowd Counting
Shenjian Gong, Shanshan Zhang, Jian Yang, Dengxin Dai, Bernt Schiele
Check 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 |
|---|---|---|---|
| apply bi-level alignment (ablation) | BLA with VGG16 · GCC → SHA | MAE: 134.7 -> 99.3 | Improved |
| add task-driven data alignment (ablation) | BLA without fine-grained feature alignment · GCC → SHA | MAE: 134.7 -> 109.1 | Improved |
| replace global feature alignment with fine-grained feature alignment (ablation) | BLA without task-driven data alignment · GCC → SHA | MAE: 134.7 -> 121.1 | Improved |
| replace task-driven data alignment with style transfer (ablation) | BLA · GCC → SHA | MAE: 99.3 -> 119.4 | Hurt |
| replace task-driven data alignment with domain randomization (ablation) | BLA · GCC → SHA | MAE: 99.3 -> 110.0 | Hurt |
| use grid size (16,16) instead of the other tested grid sizes (ablation) | fine-grained feature alignment · GCC → SHA | MAE: (2,2): 138.1; (4,4): 130.5; (8,8): 129.0; (16,16): 121.1; (32,32): 141.4 | Improved |
| use the target dataset's searched transform rather than a transform searched for another dataset (ablation) | BLA task-driven data alignment · GCC → SHA and GCC → SHB | MAE: SHA transform: SHA 99.3, SHB 14.7; SHB transform: SHA 104.3, SHB 11.9 | Improved |
| use BLA instead of the strongest reported baseline | VGG16 crowd counting · GCC → SHA | MAE: 112.4 -> 99.3 | Improved |
| use BLA instead of the strongest reported baseline | VGG16 crowd counting · GCC → SHB | MAE: 12.8 -> 11.9 | Improved |
| use BLA instead of the strongest reported baseline | VGG16 crowd counting · GCC → QNRF | MAE: 210.0 -> 198.9 | Improved |
| use BLA instead of the strongest reported baseline | VGG16 crowd counting · GCC → UCF-CC-50 | MAE: 355.0 -> 346.8 | Improved |
| use BLA instead of the strongest reported baseline | VGG16 crowd counting · GCC → WorldExpo'10 | Avg. MAE: 17.4 -> 17.9 | Hurt |
Closest papers by meaning
- Explicit Invariant Feature Induced Cross-Domain Crowd Counting (AAAI 2023 · similarity 0.89)
- Error-Aware Density Isomorphism Reconstruction for Unsupervised Cross-Domain Crowd Counting (AAAI 2021 · similarity 0.89)
- Domain-General Crowd Counting in Unseen Scenarios (AAAI 2023 · similarity 0.89)
- Single Domain Generalization for Crowd Counting (CVPR 2024 · similarity 0.87)
- Free Lunch Enhancements for Multi-modal Crowd Counting (CVPR 2025 · similarity 0.87)
- CrowdDiff: Multi-hypothesis Crowd Density Estimation using Diffusion Models (CVPR 2024 · similarity 0.86)
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.
