Bridging the Gap Between Online and In-Store Shopping: Fashion Recommendations and Virtual Try-On

The development of extensive clothing databases has significantly advanced the field of clothing recognition and recommendation systems. However, existing datasets often suffer from a limited number of annotations and face challenges when applied to diverse, real-world scenarios. To address these limitations, the proposed research work leverages DeepFashion, a comprehensive large-scale garment dataset renowned for its detailed annotations. The primary objective is to develop an advanced fashion recommendation system combined with a virtual try-on feature using cutting-edge deep learning techniques. This system is designed to enhance the online shopping experience by providing personalized clothing recommendations and enabling users to visualize how the recommended garments would appear on them. By inputting an image, users can receive recommendations of similar clothing items by using powerful feature extraction capabilities of the ResNet50 model and the efficient retrieval mechanism of the Nearest Neighbor algorithm. Following the recommendation phase, the system employs the U2Net model for image segmentation to facilitate the virtual try-on feature. U2Net's two-level layered U-structure architecture performs well in detecting salient objects, producing high-resolution segmented images that isolate the clothing item from the background and the human body in the original input. This segmentation is crucial for accurately overlaying the recommended garment onto the user's image, creating a realistic virtual try-on experience. Following the recommendation phase, the system employs the U2Net model for image segmentation to facilitate the virtual try-on feature. The proposed research aims to address the problems of current fashion recommendation systems by utilizing the extensive and richly annotated DeepFashion dataset and integrating advanced deep learning models.