DCNv4-based CLIP Text-guided Stereoscopic Image Style Transfer

The DCNv4-based CLIP text-guided stereoscopic image style transfer allows users to freely generate stereoscopic images in different styles without relying on reference images. On the one hand, this eliminates the need for large style image datasets to train the model; on the other hand, it enables the generation of dynamic stereoscopic images using only 2D images. Considering the advantages of stereoscopic images in all directions, this paper integrates the DCNv4 module into ProteusNeRF and uses CLIP text guidance for stereoscopic style transfer. A text-guided stereoscopic image style transfer framework is designed, where the adaptive capabilities of DCNv4 enhance the style transfer network’s ability to recognize and extract features of the main subject in stereoscopic images. Experimental results show that the proposed method achieves significant improvements in SSIM and PSNR metrics. Compared to the original method and other attention mechanisms, the SSIM value improves by approximately 0.2-0.5, and the PSNR increases by 1-3 dB.