Deep Learning-Powered Virtual Try-On Platforms: An Extensive Analysis of Models, Datasets, and Prospects

The rapid growth of e-commerce has transformed consumer behavior, particularly in the fashion industry, where the inability to physically try on garments remains a significant barrier to User satisfaction and purchase confidence. Virtual try-on (VTO) technologies have emerged as a promising solution to bridge the gap between online platforms and in-person shopping experiences. This review examines four principal categories of VTO systems physics-based, image-based, multi-pose, and video-based models while addressing the central research questions of how these approaches generate photorealistic, identity-preserving results and how they differ in terms of technical complexity and applicability. Deep learning techniques such as appearance flow mechanisms, diffusion models, Thin-Plate Spline (TPS) transformations, and generative adversarial networks (GANs) are analyzed, alongside segmentation and pose estimation strategies that enhance realism. The survey further compares widely used datasets, contrasts paired and unpaired training paradigms, and summarizes evaluation metrics employed to benchmark state-of-the-art models. Current challenges and research gaps are identified, and future directions for innovation are discussed. This review provides a comprehensive resource for advancing the development of VTO systems at the intersection of computer vision, fashion technology, and e-commerce