fabric-defect-detection

Built a real-time, purely classical computer vision system for fabric defect detection using multi-method analysis (GLCM, FFT, Gabor, statistical variance, background subtraction, and edge–Hough), with IoU-based bounding box fusion for robust localization. Deployed and optimized the pipeline on Jetson Nano for real time defect detection.

YOLOv8 model that detects fabric defects in images and video, trained on a benchmark defect dataset. Final-year thesis, B.Sc. Textile Engineering, BUTEX.

Native-resolution tiled YOLO11 pipeline for detecting 20 industrial fabric defects, with WBF fusion, ONNX serving, and a live project demo.

Fabric defect detection using Swin Transformer, Faster R-CNN & EfficientNet on the TILDA textile dataset.

Fine-tuned a YOLOv8 detection model on a purpose-built corpus of 720+ augmented defect samples across 4 defect classes, achieving 97% mAP@0.5 with a real-time inference pipeline deployable on commodity hardware.

FD-Net V2: A Scratch-Trained Lightweight Deep Learning Architecture for Fabric Defect Diagnosis