Research map / Point clouds / Registration and matching
Local Spacing-Aware Hungarian Matching for Stable Point-Supervised Crowd Counting
Kai Jiang, Yiming Lin, Zurui Ao
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 |
|---|---|---|---|
| replace fixed-weight Hungarian geometric cost with per-target spacing-aware SAH cost (ablation) | P2PNet-based matcher · SHHA | MAE: 50.06 -> 47.23 | Improved |
| use anchor points instead of predicted points as matching reference (ablation) | baseline matcher · SHHA | MAE: 50.89 -> 50.70 | No clear effect |
| use anchor reference with SAH rather than predicted reference (ablation) | SAH-matcher · SHHA | MAE: 51.63 -> 47.23 | Improved |
| set neighborhood size k=4 instead of other values (ablation) | SAH-matcher · SHHA | MAE: 50.10 (k=1) -> 47.23 (k=4) | Improved |
| plug SAH into P2PNet | P2PNet · SHHA | MAE: 54.70 -> 50.30 | Improved |
| plug SAH into PET | PET · SHHB | MAE: 6.56 -> 6.76 | Hurt |
| plug SAH into CLTR | CLTR · SHHA | MAE: 66.53 -> 65.20 | Improved |
| add SAH to matcher | P2PNet-based · SHHA assignment diagnostics | CAS / HR / IR: CAS 0.63 -> 0.44; HR 0.16 -> 0.02; IR 0.123 -> 0.059 | Improved |
| SAH under JHU-Crowd++ distribution shift vs APGCC | SAH-matcher · JHU-Crowd++ | MAE: 54.30 (APGCC) vs 58.32 (Ours) | Hurt |
| jitter training annotations by up to 8 px with SAH (ablation) | SAH-matcher · SHHA | MAE: +12.12 MAE vs baseline | Hurt |
| cap alpha_i at 0.08 under 8 px jitter (ablation) | SAH-matcher · SHHA (jitter 8 px) | MAE: 65.50 -> 53.14 | Improved |
Closest papers by meaning
- Rethinking Counting and Localization in Crowds: A Purely Point-Based Framework (ICCV 2021 · similarity 0.90)
- Point-Query Quadtree for Crowd Counting, Localization, and More (ICCV 2023 · similarity 0.89)
- Point-to-Region Loss for Semi-Supervised Point-Based Crowd Counting (CVPR 2025 · similarity 0.88)
- Optimal Transport Minimization: Crowd Localization on Density Maps for Semi-Supervised Counting (CVPR 2023 · similarity 0.87)
- Recurrent Attentive Zooming for Joint Crowd Counting and Precise Localization (CVPR 2019 · similarity 0.87)
- Stochastic Matching via Local Sparsification (NeurIPS 2026 · 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.
