Neural Style Transfer (NST) is an algorithm that uses neural networks to apply the style of one image to the content of another image, generating a visually appealing hybrid image. I presented this project at HackTheMist, the largest ML hackathon in Toronto, to display a well-documented implementation of the NST algorithm with Python and PyTorch, allowing the audience to merge the content (in this case, a landscape photography of Toronto) and style of distinct images (in the workshop, styles of Van Gogh, Pablo Picassso, and Salvador Dali) to create artful, AI-generated compositions.
The implementation, built on the principles laid out in Gatys et al.'s seminal paper "A Neural Algorithm of Artistic Style", uses a pre-trained VGG19 model to extract content and style features and calculate the loss function. The repository also contains diverse example outputs, tutorials, and comprehensive guides to help both novices and seasoned practitioners understand and experiment with NST.
Additionally, you'll find resources on advanced NST techniques, variations and their applications, related research papers, and links to online demonstrations. Whether you're an artist looking to incorporate AI into your toolbox, or a researcher wanting to dive deep into the world of NST, this repository has something for you.
