Pastiche Master: Exemplar-Based High-Resolution Portrait Style Transfer
Pastiche Master is a novel method for high-resolution portrait style transfer, which can generate high-quality artistic portraits from any source portrait and target style image. The method is exemplar-based, which means that it learns the style from a set of exemplar images rather than using a predefined style model.
The Pastiche Master method consists of two key components: a content-aware style transfer module and a multi-scale painting module. The style transfer module computes the style loss by comparing the Gram matrices of the feature maps extracted from the source and target images. The multi-scale painting module generates the output image by iteratively refining the coarse-to-fine details of the painting.
To train the Pastiche Master model, we use a dataset of high-quality exemplar portraits and their corresponding style images. We extract the feature maps from the VGG-19 network and compute the Gram matrices for each feature map. We then train the model by minimizing the style loss between the source image and the target style image, while preserving the content information of the source image.
To generate a high-resolution portrait painting, we first resize the source image to match the resolution of the target style image. We then pass the image through the trained Pastiche Master model to obtain the stylized image. Finally, we use the multi-scale painting module to refine the details of the painting and generate the final output.
The Pastiche Master method achieves state-of-the-art results on several benchmark datasets, including the Flickr Portrait Dataset and the WikiArt dataset. The method can generate high-quality artistic portraits that preserve the content information of the source image while transferring the style of the target image.
Overall, Pastiche Master demonstrates the effectiveness of exemplar-based style transfer for high-resolution portrait painting and opens up new opportunities for creating personalized and artistic portraits.
原文地址: http://www.cveoy.top/t/topic/nLl 著作权归作者所有。请勿转载和采集!