Konuşmacılar
Açıklama
This paper introduces an innovative project intersecting style transfer and virtual reality (VR) technologies aimed at enhancing auditory-visual experiences, particularly in the realm of music. Our dynamic platform empowers musicians, spanning from novices to experts, to tailor their compositions within engaging environments enriched with stylized visual and auditory elements. Using deep learning methodologies such as Traditional Neural Style Transfer (TNST) and Neural Neighbor Style Transfer (NNST) in conjunction with VR technology, our framework enables users to craft personalized experiences aligned with their preferences. In this paper, we demonstrate a hybrid approach to style transfer that integrates NNST with conventional techniques and leverages multi-GPU processing for enhanced efficiency. Despite inducing a 19% slowdown compared to NNST alone, our user survey confirmed the enhanced realism of images generated via the hybrid method. Furthermore, we have achieved limited acceleration through multi-GPU support. In conclusion, we have developed a hybrid style transfer method from 2D to 2D and seamlessly integrated it into our VR environment. The envisioned extension of style transfer from 2D to 3D necessitated adaptation to overlay 2D images onto 3D objects due to the processing limitations of VR goggles.