Description

WorldExpo '10 (Zhang et al.) is a video crowd-counting dataset captured at the 2010 Shanghai World Expo across multiple scenes and cameras. Each test scene contains 120 labelled frames with crowd counts ranging from a single pedestrian up to roughly 220 per frame. It is the standard cross-scene benchmark for crowd counting under camera and viewpoint diversity, and is regularly criticised as insufficient for extremely dense crowds.

Modality / size

  • Modality: RGB video frames from multiple surveillance cameras.
  • Subjects / scenarios: 5 test scenes, 120 labelled frames each; multiple training scenes.
  • Labels: per-pedestrian head annotations.

Used by (papers)

  • Standard cross-scene baseline; reported next to ShanghaiTech, UCF_CC_50, UCF-QNRF, Mall.

Notes

  • Public dataset accompanying the WorldExpo crowd-counting paper.

3 vault papers evaluate on this dataset

Titles and DOIs only — no abstracts, no analyses.

  • A survey of recent advances in CNN-based single image crowd counting and density estimation 2018 DOI ↗
  • Data-driven Crowd Modeling Techniques: A Survey 2022 DOI ↗
  • Constructing WiFi-Video-Fused Multi-Modal Synthetic Datasets for Crowd Counting 2025 DOI ↗