Derwya/half-life_player_seg

Half-Life Player Segmentation Dataset

This dataset contains in-game screenshots from Half-Life, annotated with polygon masks for player models. It is specifically designed to train YOLO segmentation models for real-time player tracking and aim assistance algorithms.

Dataset Details

• Format: YOLO Segmentation (txt files with normalized polygon coordinates)
• Total Images: 598 (balanced to include empty background frames)
• Splits:
  • Train: 539 images
  • Validation: 59 images
• Classes: 1 class (0: player)

Structure

The dataset follows the standard YOLO format:

• images/train/ & images/val/: Images captured directly from gameplay.
• labels/train/ & labels/val/: Text files containing the segmentation mask coordinates for each image.
• data.yaml: The configuration file required to start YOLO training.

Intended Use

This dataset is intended for computer vision research in gaming environments. By using segmentation masks instead of standard bounding boxes, models trained on this dataset can locate the exact outline of a player. This enables much higher precision for aiming algorithms, reducing false lock-ons on complex backgrounds and allowing trackers to calculate true center-of-mass or head positions.