This repo provides a YOLOv8 model, finely trained for detecting human heads in complex crowd scenes, with the CrowdHuman dataset serving as training data. To boost accessibility and compatibility, I've reconstructed the labels in the CrowdHuman dataset, removing useless annotations, refining its annotations to perfectly match the YOLO format.

Download the CrowdDataset from [https://www.crowdhuman.org/download.html]. I have provided the YOLO format labels in the CrowdHumanHead/labels.zip file. Simply unzip this file and place the contents in the CrowdHumanHead/labels directory.

Please download the model weight from this Google Drive URL.

python3 pred_yolo_v8.py

python3 train_yolo_v8.py

Scrip/create_chuman.py -- Creat txt format labels from ODGT. Scrip/gen_labels.py -- Generate YOLO labels from txt format. Scrip/vis_labels.py-- Visualize YOLO labels for label checking.