m monad-knowledge FIIT STU · open research vault
dataset

shanghaitech

Also known as: ShanghaiTech, ShanghaiTech dataset, ShanghaiTech A, ShanghaiTech B, ShanghaiTech A/B, Shanghai Tech

Description

ShanghaiTech (Zhang et al.) is a large-scale crowd-counting image dataset split into two parts: Part A consists of crowded internet images with high density variability, while Part B consists of street-view images from Shanghai with sparser crowds. It is one of the most widely used crowd-counting benchmarks and is reported in nearly every modern density-estimation paper.

Modality / size

  • Modality: still RGB images with per-pedestrian head annotations.
  • Subjects / scenarios: Part A (internet, dense) and Part B (street view, sparser).
  • Labels: head annotations and density maps.
  • Known limitation: many low-density samples, few extremely high-density samples.

Used by (papers)

  • Cited next to UCF_CC_50, UCF-QNRF, WorldExpo '10, and Mall in crowd-counting benchmark tables.
  • Used as the video baseline in CSI-vs-vision comparison studies.

Notes

  • Public release with the MCNN paper.
Where it appears

2 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 ↗
  • Constructing WiFi-Video-Fused Multi-Modal Synthetic Datasets for Crowd Counting 2025 DOI ↗
Co-mentioned with
UCSD×2 WorldExpo '10×2 Domain Adaptation×1 Transformer / Attention Mechanism×1 Fine-Tuning×1 Pedestrian Dynamics Simulation×1 ImageNet×1 Deep Learning×1 Neural Network×1 Crowd Density Estimation×1