11.2K ·GitHub
What is this?
FastPhotoStyle is NVIDIA's photorealistic style transfer code: given a content photo and a style photo, it transfers the style of the style photo onto the content photo using a closed-form algorithm from an ECCV 2018 paper.
FastPhotoStyle is NVIDIA's photorealistic style transfer code: given a content photo and a style photo, it transfers the style of the style photo onto the content photo using a closed-form algorithm from an ECCV 2018 paper.Is this a polished app, or research code I run myself?It's a research code release accompanying an academic paper, not a packaged consumer app — you run the provided scripts/tutorial yourself on PyTorch.
It's a research code release accompanying an academic paper, not a packaged consumer app — you run the provided scripts/tutorial yourself on PyTorch.How do I try it out?Follow the included tutorial, which shows three ways of using the FastPhotoStyle algorithm; a Docker setup is also provided.
Follow the included tutorial, which shows three ways of using the FastPhotoStyle algorithm; a Docker setup is also provided.Which PyTorch version do I need?The current code targets PyTorch 0.4.0; if you're on PyTorch 0.3.0 there's a separate branch for that version.
The current code targets PyTorch 0.4.0; if you're on PyTorch 0.3.0 there's a separate branch for that version.Can I use this commercially?No — the code is licensed under CC BY-NC-SA 4.0, which restricts it to non-commercial use with share-alike requirements.
No — the code is licensed under CC BY-NC-SA 4.0, which restricts it to non-commercial use with share-alike requirements.What paper is this based on?It implements 'A Closed-form Solution to Photorealistic Image Stylization' by Yijun Li, Ming-Yu Liu, Xueting Li, Ming-Hsuan Yang, and Jan Kautz, presented at ECCV 2018.
It implements 'A Closed-form Solution to Photorealistic Image Stylization' by Yijun Li, Ming-Yu Liu, Xueting Li, Ming-Hsuan Yang, and Jan Kautz, presented at ECCV 2018.Why does the closed-form approach matter?Instead of iterative optimization, the method uses a closed-form feature transform, making photorealistic style transfer faster and more exact while avoiding typical style-transfer artifacts.
Instead of iterative optimization, the method uses a closed-form feature transform, making photorealistic style transfer faster and more exact while avoiding typical style-transfer artifacts.Where should I start if I'm new to this repo?Start with the provided tutorial, which walks through three concrete ways to run the FastPhotoStyle algorithm on your own content/style photo pairs.
Start with the provided tutorial, which walks through three concrete ways to run the FastPhotoStyle algorithm on your own content/style photo pairs.What are similar projects?object-permanence, PyTorch-GAN, Awesome-pytorch-list, CycleGAN, pix2pix, Attention-Residuals, Time-Series-Library, pytorch-CycleGAN-and-pix2pix
object-permanence, PyTorch-GAN, Awesome-pytorch-list, CycleGAN, pix2pix, Attention-Residuals, Time-Series-Library, pytorch-CycleGAN-and-pix2pixWhich topics is it in?style-transfer, cv, arxiv, papers, image, neural-network, academic-research, image-processing, artificial-intelligence, nvidia, computer-vision, pytorch
style-transfer, cv, arxiv, papers, image, neural-network, academic-research, image-processing, artificial-intelligence, nvidia, computer-vision, pytorch