Virtual try-on technology has seen significant advancements in recent years, particularly in the domain of apparel visualization. However, a notable gap exists in high-resolution (720 × 540) datasets that adequately represent diverse poses and orientations for Indic clothing styles. To address this, we introduce a novel dataset, IndicViton, comprising multiple poses of the same garment around the wearer. This dataset fills a critical void as existing open-source collections lack such diversity. In addition to dataset creation, we propose an initial diffusion model (DIVA) tailored for Indic virtual try-on applications. Our model leverages a novel approach to handle multi-pose garment images, enabling realistic and accurate virtual fitting experiences. Central to our methodology is the utilization of diffusion models, which excel in capturing intricate details and variations in fabric textures.
To validate our approach, we conducted comprehensive experimental evaluations, comparing our results against established benchmarks. Results demonstrated superior visual fidelity and quantitative performance in Indic virtual try-on scenarios. By providing this dataset and model, we aim to spur further research and development in virtual try-on technologies tailored for Indic clothing styles. The link for the dataset is attached here.
