CrowdHuman Dataset Parsing

This repo contains script to read the crowd human dataset in a structured format

Getting Started

python3 parseDataset.py --config_file_path=./config.txt --data test

How to define config file for your requirements

[dataset] <br /> dataset_base_dir path to directory where dataset folders are present after downloading <br /> train_annotation_file_path path to train annotation file <br /> test_annotation_file_path path to test annotation file <br />

<br />

[box_attributes] <br /> visible=1 will allow visible bbox. <br /> full=1 will allow full box. <br /> head=1 will allow head box <br /> occ=1 will allow occluded bboxes. <br /> unsure=0 will not allow unsure bboxes. <br /> h_ignore=0 will not allow ignored region bboxes. <br /> h_occ=0 will not allow occluded heads. <br /> h_unsure=0 will not allow unsure head bboxes. <br /> tag=person will allow only bboxes with tag of person (other is mask). <br />

<br />

[test] <br /> debug=1 to debug by plotting bbox on below mentioned 'num_image' random images <br /> num_images=5 <br />

The dataset download link is mentioned in the website mentioned below.

Crowd Human Dataset Website

Crowd Human Dataset Paper Link

Requirements

1. OpenCV : pip install opencv-python