Multi-class Real-Time Fabric Object Detection via YOLO

Built for a Tsing Hua University masteral student researching automated quality control in textile manufacturing. The system uses YOLOv8 to detect and classify multiple fabric defect types in real time — 94% accuracy at 45ms per inference after ONNX optimization. Training data was managed through Roboflow, and the final system ships with a Streamlit admin interface for uploading footage and reviewing detections. Reduced manual inspection time by 60% in testing conditions.