Virtual top try-on systems have revolutionized online shopping, allowing consumers to visualize clothing items before purchasing. This paper presents a novel approach to virtual top try-on, integrating attribute-controlled and Geometry-Preserving GAN (ACGPN), U-2 Net architecture, and human parsing. ACGPN facilitates the generation of realistic top simulations with customizable attributes, while the U-2 Net architecture enhances segmentation accuracy for precise garment placement. Human parsing ensures accurate detection and segmentation of body parts, optimizing virtual garment fitting. Experimental results demonstrate the effectiveness of the proposed approach in achieving lifelike virtual try-on experiences. The model is verified on the VITON dataset and the Custom dataset.

Virtual top try-on, Attribute-Controlled and Geometry-Preserving GAN (ACGPN), U-2 Net architecture, Human parsing.