Research map / Label-efficient and robust learning / Domain adaptation and generalization
Exploring Scale Shift in Crowd Localization under the Context of Domain Generalization
Juncheng Wang, Lei Shang, Ziqi Liu, Wang Lu, Xixu Hu, Zhe Hu, Jindong Wang, Shujun Wang
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 SemanticHook (ablation) | ERM · ScaleBench avg | F1: 77.47 -> 78.60 | Improved |
| add ScaledLoss on top of SemanticHook (ablation) | Catto · ScaleBench avg | F1: 78.60 -> 79.27 | Improved |
| add S2InvLoss on top of SemanticHook (ablation) | Catto · ScaleBench avg | F1: 78.60 -> 78.63 | No clear effect |
| add S2InvLoss and ScaledLoss together (ablation) | Catto with SemanticHook · ScaleBench avg | F1: 78.60 -> 79.69 | Improved |
| use Gaussian perturbation rather than interpolation (scale) perturbation (ablation) | SemanticHook · ScaleBench avg | F1: 70.90 -> 78.60 | Improved |
| use ColorJitter rather than Gaussian perturbation (ablation) | SemanticHook · ScaleBench avg | F1: 78.60 vs 78.50 | No clear effect |
| enhance hooked semantic feature rather than global feature (ablation) | Catto · ScaleBench avg | F1: 69.13 -> 78.60 | Improved |
| add annealing of gamma (ablation) | Catto · ScaleBench avg | F1: 73.03 -> 78.60 | Improved |
| random interpolation augmentation (RA) in training (ablation) | Big-source model · Big->smaller domains avg | F1: 56.14 -> 57.96 | Mixed |
| inference-time interpolation (IA) (ablation) | Big-source model · Big->smaller domains avg | F1: 56.14 -> 53.01 | Hurt |
| add domain B to training for target S (ablation) | multi-source training · ScaleBench target Small | F1: 77.92 -> 77.94 | No clear effect |
| use 30% of InD data instead of full (ablation) | crowd locator · InD scale split | F1 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Coral — CORAL: Disentangling Latent Representations in Long-Tailed Diffusion (NeurIPS 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CrowdDriven: A New Challenging Dataset for Outdoor Visual Localization (ICCV 2021 · similarity 0.88)
- Localization in the Crowd with Topological Constraints (AAAI 2021 · similarity 0.87)
- Rethinking Counting and Localization in Crowds: A Purely Point-Based Framework (ICCV 2021 · similarity 0.87)
- Counting without Scale: Scale-Consistent Error Correction for Crowd Counting (NeurIPS 2026 · similarity 0.87)
- AutoScale: Learning to Scale for Crowd Counting (IJCV 2022 · similarity 0.86)
- Point-to-Region Loss for Semi-Supervised Point-Based Crowd Counting (CVPR 2025 · similarity 0.85)
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.
