This repo contains script to read the crowd human dataset in a structured format
python3 parseDataset.py --config_file_path=./config.txt --data test
[dataset]
dataset_base_dir path to directory where dataset folders are present after downloading
train_annotation_file_path path to train annotation file
test_annotation_file_path path to test annotation file
[box_attributes]
visible=1 will allow visible bbox.
full=1 will allow full box.
head=1 will allow head box
occ=1 will allow occluded bboxes.
unsure=0 will not allow unsure bboxes.
h_ignore=0 will not allow ignored region bboxes.
h_occ=0 will not allow occluded heads.
h_unsure=0 will not allow unsure head bboxes.
tag=person will allow only bboxes with tag of person (other is mask).
[test]
debug=1 to debug by plotting bbox on below mentioned 'num_image' random images
num_images=5
The dataset download link is mentioned in the website mentioned below.
Crowd Human Dataset Paper Link
- OpenCV :
pip install opencv-python