object detection for autonomous car

This project presents a cleaned, enhanced, and YOLO-ready version of the popular BDD100K traffic dataset, tailored for high-performance object detection models.

Key Enhancements & Processing

Added higher-resolution images to improve the quality and balance of traffic-related object classes.

Focused on six critical classes:

train, bike, motor, rider, traffic sign, and traffic light.

Removed:

All bounding boxes smaller than 10% of the average class size.

Duplicate images and inconsistent labels.

Samples with missing images or annotations.

Verified YOLO-format compatibility and annotation integrity.

Trained yolov11s achieved map 50 : 82%

dataset size 115k images

tested on realtime problems in different cases