Project: BlockBlast Solver — manual screenshot labeling for vision model training
Label Studio
Image
Object DetectionManually labeled 120 game screenshots using Label Studio to train a vision model for board detection. The labels were used alongside YOLO-based board detection and a PPO-trained reinforcement learning policy for selecting optimal moves. This activity supported development of a vision pipeline that converts screenshots into structured board state information for downstream decision-making.• Labeled 120 screenshots in Label Studio for model training• Supported YOLO board detection dataset preparation• Integrated labeled outputs into a larger inference and move-selection pipeline• Wrote unit tests for core rules, environment behavior, and vision edge cases
2020