Ai Based Virtual Try On System Using Dl

• Author(s):

  Saurabh Brahmankar | Sagar Arjun Kharat | Vishal Raghunath Dipake | Aditya Arvind Kawade | Prof. S. B. Nimbekar

• Keywords:

  Virtual Try-On, Conditional GAN, Geometric Matching Module, Thin-Plate Spline, MERN Stack, Deep Learning, Image Synthesis, Pose Estimation.

• Abstract:

  The Proliferation Of E-commerce Has Fundamentally Altered Consumer Purchasing Behaviour, Yet A Persistent Challenge Remains: The Inability To Physically Try On Garments Before Purchase. This Paper Presents A Virtual Clothing Try-On (VCTO) System That Bridges This Gap By Combining A Custom Conditional Generative Adversarial Network (cGAN) With A Geometric Matching Module (GMM) Embedded In A Full-stack MERN (MongoDB, Express.js, React.js, Node.js) Web Application. The Proposed System Accepts A Reference Person Image And A Desired Garment Image As Inputs And Synthesises A Photorealistic Composite Image By (i) Estimating Human Body Pose Using A Lightweight Keypoint Detector, (ii) Warping The Garment Via Thin-Plate Spline (TPS) Transformation, And (iii) Generating The Final Try-on Image Through An Adversarial Training Scheme. Experimental Evaluation On The VITON-HD Benchmark Dataset Yields A Structural Similarity Index (SSIM) Of 0.873, Fréchet Inception Distance (FID) Of 8.34, And Learned Perceptual Image Patch Similarity (LPIPS) Of 0.072, Outperforming Several Baseline GAN-based Methods. The System Achieves An Average Inference Latency Of 320 Ms On A Single NVIDIA RTX 3060 GPU, Making It Suitable For Near-real-time Web Deployment.

---

Other Details

• Paper id:

  IJSARTV12I6105604

• Published in:

  Volume: 12 Issue: 6 June 2026

• Publication Date:

  2026-06-04

Download Article