Research map / Detection and segmentation / CNN features
Consistency-Aware Anchor Pyramid Network for Crowd Localization
Xinyan Liu, Guorong Li, Yuankai Qi, Zhenjun Han, Anton van den Hengel, Nicu Sebe, Ming–Hsuan Yang, Qingming Huang
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 AAPS and CTR together (ablation) | anchor-based crowd localizer · JHU-CROWD++ | F1: 76.2 -> 82.5 | Improved |
| add CTR to the baseline without AAPS (ablation) | anchor-based crowd localizer · JHU-CROWD++ | F1: 76.2 -> 78.7 | Improved |
| add AAPS to the baseline without CTR (ablation) | anchor-based crowd localizer · JHU-CROWD++ | F1: 76.2 -> 81.5 | Improved |
| add CTR on top of AAPS (ablation) | anchor-based crowd localizer · JHU-CROWD++ | F1: 81.5 -> 82.5 | Improved |
| use three anchor levels 1,4,8 instead of single level of 4 anchors (ablation) | AAPS · JHU-CROWD++ | F1: 78.7 -> 82.5 | Improved |
| use anchor counts 1,2,4 instead of 1,4,8 (ablation) | AAPS · JHU-CROWD++ | F1: 82.5 -> 76.2 | Hurt |
| add a fourth anchor level (10 to 16 anchors) (ablation) | AAPS · JHU-CROWD++ | F1: 82.5 -> 82.1 | No clear effect |
| add K-means anchor priors instead of uniform anchors (ablation) | AAPS · JHU-CROWD++ | F1: 82.1 -> 82.5 | Improved |
| add cascade region count loss instead of MSE loss (ablation) | AAPS · JHU-CROWD++ | F1: 80.8 -> 82.5 | Improved |
| add cascade region count loss instead of MSE loss (ablation) | AAPS · JHU-CROWD++ | MAE: 63.4 vs 64.2 | Hurt |
| use ConvNext-S instead of HRNet-48 backbone | proposed method · JHU-CROWD++ | F1: 80.7 -> 82.5 | Improved |
| use ConvNext-S instead of HRNet-48 backbone | proposed method · UCF-QNRF | F1: 79.3 vs 79.2 | No clear effect |
Closest papers by meaning
- Rethinking Counting and Localization in Crowds: A Purely Point-Based Framework (ICCV 2021 · similarity 0.87)
- Exploring Scale Shift in Crowd Localization under the Context of Domain Generalization (IJCV 2026 · similarity 0.85)
- Recurrent Attentive Zooming for Joint Crowd Counting and Precise Localization (CVPR 2019 · similarity 0.85)
- An End-to-End Transformer Model for Crowd Localization (ECCV 2022 · similarity 0.85)
- Locate, Size and Count: Accurately Resolving People in Dense Crowds via Detection (TPAMI 2020 · similarity 0.85)
- Local Spacing-Aware Hungarian Matching for Stable Point-Supervised Crowd Counting (ECCV 2026 · 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.
