Dataset Comparison

Owned by Shrinjay Mukherjee
Last updated: 2021-10-08 by Jinghao Lei
1 min read

UAV Human

Page: https://github.com/SUTDCV/UAV-Human

Download Link: https://drive.google.com/drive/folders/1QeYXeM_pbWBSSmpRr_rKHurMpI2TxAKs

A dataset containing a wider variety of traffic scenarios than the Stanford set. The angle of the camera is more of a bird's eye and a bigger focus on human behaviour rather than just the detection of a person. Similar to the Stanford set, it is also videos and is 67,428 videos long.

The videos appeared to be unlabelled however, the repo can convert the videos to frames and then each frame we will draw bounding boxes and then add Yolo format labels

 

Stanford

Page: https://cvgl.stanford.edu/projects/uav_data/ (Download is on the site) (69GB zip file of videos)

A dataset containing videos of traffic. Each separate entity is labelled with colored boxes

The dataset labels bicyclists, pedestrians, skateboarders, carts, cars and buses. (Honestly if our goal is to detect people, it might not be very helpful to have all those different types of traffic as well. The angle of the video is top down

Pedestrians and objects are labelled with colored bounding boxes however there is no pixel location labelling for entities

This repo https://github.com/rockkingjy/DataFormat_sdd2kitti is able to convert the videos to images and then make position of bounding boxes to a text file in KITTI format
This repo https://github.com/CUFCTL/DLBD/blob/master/scripts/convert-kitti-to-yolo.py can then convert KITTI format to YOLO format

UCF-ARG

… details go here

Possible other datasets

…details go here