GarmentIQ: Automated Garment Measurement for Fashion Retail

Online fashion retail has revolutionized shopping but faces persistent sizing issues, driving return rates above 25%, costing fashion retailers billions of dollar annually, and increasing textile waste and carbon emissions. We present GarmentIQ, an end-to-end computer vision system combining garment classification, high-resolution segmentation, and landmark detection for precise, template-free measurements across nine categories. An interactive web interface lets users define custom measurement points and export structured JSON and PDF instructions. We used a 23,266-image dataset from Nordstrom and Myntra, and our tinyViT classifier achieves 95.76% accuracy, demonstrating superior generalization after fine-tuning on Zara data. BiRefNet produces high-quality segmentation, and HRNet-based landmark extraction attains high precision, with customized landmark derivation. GarmentIQ's modular, user-friendly design streamlines workflows, reduces returns, and promotes sustainability, laying the groundwork for future automated fashion analysis.