ucf-cc-50

Description

UCF_CC_50 (Idrees et al.) is an image-based crowd-counting dataset of 50 challenging high-density crowd images sourced from the web. It was the first benchmark designed explicitly to cover a wide range of densities and diverse scenes with strong perspective distortion, and is typically evaluated with five-fold cross-validation due to its small size.

Modality / size

• Modality: still RGB images with per-head annotations.
• Subjects / scenarios: 50 images of dense crowds (concerts, protests, stadiums, religious gatherings).
• Labels: head dot annotations and density maps.

Used by (papers)

• Standard small-but-dense benchmark, almost always paired with ShanghaiTech and WorldExpo '10.
• Five-fold cross-validation protocol is the convention.

Notes

• Public dataset; UCF Computer Vision lab release.