Research map / Learning theory and algorithms / Clustering
Mahalanobis Distance-based Multi-view Optimal Transport for Multi-view Crowd Localization
Qi Zhang, Kaiyi Zhang, Antoni B. Chan, Hui 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 |
|---|---|---|---|
| replace Gaussian density MSE loss with Euclidean multi-view optimal transport loss (ablation) | multi-view crowd localization on CVCS · CVCS | MODA: 18.6 -> 42.8 | Improved |
| add view-ray direction and object-to-camera distance to the Mahalanobis cost (M-MVOT vs E-MVOT) (ablation) | E-MVOT · CVCS | MODA: 42.8 -> 45.8 | Improved |
| add object-to-camera distance only (ED-MVOT) to Euclidean OT (ablation) | E-MVOT · CVCS | MODA: 42.8 -> 43.8 | Improved |
| M-MVOT instead of E-MVOT on MultiviewX (ablation) | MVDeTr backbone · MultiviewX | MODA: 96.3 -> 96.7 | Improved |
| penalize deviations along view ray more (sigma1^2 < sigma2^2) versus the reverse (ablation) | MV-MVOT · MultiviewX | MODP: 84.6 (1.2:1) -> 85.6 (1:1.2) | Improved |
| select closest camera for each point instead of averaging camera losses (ablation) | M-MVOT · MultiviewX | MODA: 96.5 -> 96.7 | Improved |
| replace geometric covariance with learnable covariance matrix (ablation) | M-MVOT · MultiviewX | MODA: 96.7 -> 95.1 | Hurt |
| use single-image UOT with simple multi-view sum fusion | multi-view localization · MultiviewX | MODA: 96.7 -> 23.9 | Hurt |
| replace MSE or Focal loss with M-MVOT (ablation) | MVDeTr backbone · MultiviewX | MODA: 93.9 (MSE) / 94.5 (Focal) -> 96.7 | Improved |
| increase alpha from 0.05 to 0.6 (ablation) | M-MVOT · MultiviewX | MODA: 96.7 -> 96.3 | Hurt |
| raise tau from 20 to 30 (ablation) | M-MVOT · MultiviewX | MODA: 96.7 -> 95.2 | Hurt |
| use M-MVOT instead of 3DROM on Wildtrack | multi-view localization · Wildtrack | MODA: 93.5 (3DROM) vs 92.1 | Hurt |
Closest papers by meaning
- Multi-View Optimal Transport for Multi-View Crowd Localization and Tracking (IJCV 2026 · similarity 0.95)
- SynMVCrowd: A Large Synthetic Benchmark for Multi-view Crowd Counting and Localization (IJCV 2026 · similarity 0.83)
- A Generalized Loss Function for Crowd Counting and Localization (CVPR 2021 · similarity 0.83)
- Learning to Count via Unbalanced Optimal Transport (AAAI 2021 · similarity 0.82)
- Rethinking Counting and Localization in Crowds: A Purely Point-Based Framework (ICCV 2021 · similarity 0.81)
- Multi-View 3D Human Tracking in Crowded Scenes (AAAI 2016 · similarity 0.81)
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.
