Deep Learning-Driven Virtual Clothing Fit Detection and Automatic Pattern Correction Technology

Online fashion platforms continue to experience significant refund rates due to inaccurate virtual fit assessments and the lack of integrated, automatic Fabric Pattern Adjustment. Existing virtual try-on methods typically rely on 2D image alignment and mesh-based simulations, which provide poor collision awareness and inefficient large-scale deformation modelling. To solve the above issue, this Work introduces a unified Deep Learning Framework (SDF-FNO) that combines a Signed Distance Function (SDF)-based neural network for virtual clothing fit identification and a Fourier Neural Operator (FNO) for automatic garment pattern correction. The SDF network uses collision-aware surface evaluation to determine tight, normal, and loose fit zones by modelling the human body and garment as continuous implicit surfaces and computing signed distance differences. The FNO learns to modify the entire garment pattern at once, then reconstructs a smooth, corrected pattern by analysing global shape changes in the frequency domain. The system is trained and tested using the 4D-DRESS dataset, which contains dynamic 3D clothed human scans. The SDF-FNO demonstrates scalability across e-commerce virtual try-on, augmented-reality retail, digital tailoring, and intelligent clothing production systems. The experimental results show 94.6% fit classification accuracy, 93.9% automated pattern correction success rate, and 94.3% overall application-level accuracy, while maintaining stable real-time inference performance.