Skip to content

Latest commit

 

History

History
66 lines (36 loc) · 3.15 KB

Datasets.md

File metadata and controls

66 lines (36 loc) · 3.15 KB

Dataset Descriptions

A copy of all Datasets used is stored in a pubic S3 repository for easy replication of the results and analysis.

Training

UCF-QNRF, 2018

Largest Crowd annotated dataset till date, made available for crowd counting and localization techniques by Center for Research in Computer Vision, University of Central Florida.

Details:

Number of Images Number of Annotations Average Count Maximum Count Average Resolution Average Density
1,535 (Train-1201 Test-334) 1,251,642 815 12,865 2013 x 2902 1.12 x 10-4

*The average density, i.e., the number of people per pixel over all images is also the lowest, signifying high-quality large images.

  • All images are taken from Flickr,Web and Hajj footage (Research Paper Section 4 - Data Collection) and not from a surveleilance camera streams or simulated crowd scenes. Hence, it is very diverse in terms of prepectivity, image resolution, crowd density and the scenarios which a crowd exist.

  • Dataset contains buildings, vegetation, sky and roads as they are present in realistic scenarios captured in the wild. This makes this dataset more realistic as well as difficult.

  • Reduce geographical bias images have been taken from various countries. Refer link for geo-tagged map.

UCF-CC-50, 2013

50 High Density Crowd images sourced from Flickr for research purposes by Center for Research in Computer Vision, University of Central Florida.

Details:

Number of Images Number of Annotations Average Count Maximum Count Average Resolution Average Density
50 63,974 1,279 4633 2101 x 2888 2.02 x 10-4

ShanghaiTech Dataset, 2016

[Part A](Add S3 Link) - Randomly crawled from the web.

[Part B](Add S3 Link) - Taken from busy streets of metropolitan areas in Shanghai.

Details:

Dataset Number of Images Number of Annotations Average Count Maximum Count Average Resolution Average Density
Part A 482 (1198 Total) 241,677 (330,000 Total) 501.4 3139 589 x 868 9.33 x 10-4
Part B 716 (1198 Total) 88,488 (330,000 Total) 123.6 578 768 × 1024 much larger

WorldExpo'10 Dataset, 2015

Details:

Number of Images Number of Annotations Average Count Maximum Count Average Resolution Average Density
3980 225,216 56 334 576 x 720 1.36 x 10-4
  • It includes 1132 annotated video sequences captured by 108 surveillance cameras, all from Shanghai 2010 WorldExpo (Data shared for research purposes).
  • Data was primarily generated to show effective cross-scene counting by using CNN models and density maps.

Testing

Pedestrian Dynamics Data Archive