Vision Fit: An Ai-powered Virtual Fitting Assistant For Personalized Clothing Size Recommendation Using Dual-anchor Anthropometric Calibration

Abstract

There Has Been A Significant Shift In Clothing Shopping Habits Over The Past Decade, With Most Purchases Now Made Online. A Major Challenge Remains Consumers' Inability To Try On Products Before Purchase, Leading To Return Rates Of 30–40% Due To Incorrect Sizing. This Paper Presents Vision Fit, A System That Resolves This Problem Using An Ordinary Laptop Webcam. Vision Fit Employs A Dual-Anchor Anthropometric Calibration Pipeline Using Blaze Pose To Detect 33 Body Landmarks, Then Converts Pixel Coordinates Into Centimetres Without Physical Reference Objects. Two Anatomical Proportionality Ratios (Nose-Hip ≈ 48% And Nose-Ankle ≈ 82% Of Standing Height) Are Fused With A 0.6:0.4 Weight To Derive A Robust Scale Factor. A 30-frame Temporal Stabilization Pipeline With Jitter Rejection Reduces RMSE From 2.15 Cm To 0.71 Cm — A 67% Improvement. The Brand Advisory Module Achieves 94% Size-label Accuracy Across 10 Major Apparel Brands With No Missed Recommendations Over 50 Subjects. The System Operates Fully Offline Via FastAPI And PyWebView.

Keywords