Garment-Specific Style Transfer Using VGG19 and Semantic Segmentation for High-Fidelity Clothing Image Generation

This paper will present a new style transfer technique, applied specifically to clothing images, by using the VGG19 neural network together with a semantic segmentation approach. The process involves the creation of a segmenting network where the clothing is properly separated from the background such that this style transfer technique may be more specific and targeted. This targeted approach captures the important garment features, including textures and boundary details of the garments, during the transformation. To further enhance the quality of the results, it introduces loss functions custom-designed to preserve the smoothness of the fabric and the integrity of edges of the garment. In this method, proposed here, through immense experimentation, the performance in output is superbly better than that of conventionally used techniques as the quality of the output visuals will be more accurate with all the structural elements of the garments.