Abstract: "What if we can bring our imagination into reality?" This is something this paper is trying to address with the help of image style transfer and CLIP. Image Style transfer is a technique used to take two images- a content image and a style reference image and blend them together so the output image looks like the content image, but “painted” in the style of the style reference image. As stated above, existing style transfer methods require reference style images to transfer texture information to content images. But what if one does not have a reference image but is still interested in transferring styles by just imagining them? To overcome the above issue, here we are using text-to-image embedding model of CLIP to perform text-driven image style transfer without the use of a style image as a reference. We also use a CNN Encoder-Decoder Model to capture the visual features of the content image and simultaneously stylize the image to obtain a realistic texture representation. Finally, the above model is deployed as a web application from which users can perform style transfer by uploading an image and specifying the desired styling with the help of a text description.
Keywords: Style Transfer, CNN, CLIP, VGG, U-Net.
|
DOI:
10.17148/IJARCCE.2022.11543
