Annotated Image Dataset for Fenit: Automatic Chessboard Digitization and FEN Generation

Description

This collection features a variety of chessboard states captured under different conditions. To ensure a robust model, the dataset provides a comprehensive variety of lighting, angles, and piece positioning.

Steps to reproduce

The raw source images for the dataset were annotated utilizing the Computer Vision Annotation Tool (CVAT). To expedite the labeling of complex multi-class piece bounding boxes and four-point board keypoint geometries, an automated annotation pipeline was developed utilizing the CVAT. A pre-trained base pose-estimation model (yolov8l-pose.pt) was deployed to infer preliminary bounding box boundaries and keypoint coordinates across the unlabeled set.

Institutions

• Binus University

  Jakarta, Jakarta

Categories

Licence