Comparative Analysis of CNN Models for Fabric Defect Classification

Companion research to the YOLOv8 detection project, also for a Tsing Hua University masteral student. This one is a systematic comparison — evaluating multiple CNN models from TensorFlow Model Garden using transfer learning on a custom fabric defect dataset. Tested architectures include ResNet and MobileNet variants, measuring classification performance improvements over a baseline model. The fine-tuned models improved classification by 23% over baseline. The goal was to give the client solid evidence for which architecture to commit to in their broader research.