AI Body Measurement System

High-precision fashion body measurement prediction system using deep learning. Automatically estimates 7 critical body measurements from just 2 RGB photos (frontal + lateral) plus height — no weight input required. Uses BMnet architecture with MNASNet1.0 backbone (3.27M parameters).

Achieves 98.2% accuracy on validation and 96.4% on unseen data. Trained 200 epochs on RTX 4060.

Key Features

• Predicts 7 measurements: arm-length, chest, hip, leg-length, shoulder-breadth, shoulder-to-crotch, waist
• Only requires 2 photos (front + side) and height
• 98.2% validation accuracy / 96.4% test accuracy
• BMnet deep learning model with MNASNet1.0 backbone
• No weight input required
• GPU-accelerated inference (CUDA 12.1+)
• Annotated visualization output showing measurements

Project Gallery