What it does This repo is the companion to a 2019 IEEE TVCG review paper. It catalogs selected neural style transfer papers, links to their code (Torch, TensorFlow, PyTorch, Caffe, Chainer, MXNet, MatConvNet, Theano), and hosts pre-trained models and comparison images. The authors also invite pull requests for missing papers, which keeps the list from fossilizing.

The interesting bit The taxonomy is the real product. Papers are sorted by whether they optimize images online or models offline, then subdivided by how many styles a model can handle (per-style, multi-style, arbitrary). This turns a scattered literature into something you can actually navigate. The repo also bundles supplementary materials and pre-trained models from the original paper, which is rarer than it should be in academic repositories.

Key highlights

• Covers foundational work (Gatys et al.) through 2020 methods like Dynamic Instance Normalization
• Links to implementations across seven deep-learning frameworks, including now-deprecated ones (Torch, Theano, Chainer)
• Provides pre-trained models, content/style images, and stylized results for direct comparison
• Authors explicitly welcome corrections and additions via email or PR
• Companion OSF repository for additional materials

Caveats

• Several code links point to frameworks that are effectively dead (Torch, Caffe, Theano)
• The README is a flat list; there’s no search or filtering beyond the manual taxonomy
• Last major update appears to be 2019–2020; newer methods may be missing

Verdict Worth bookmarking if you’re doing a literature review, reproducing classic results, or teaching the topic. Skip it if you want a unified framework—look at the linked pystiche project instead.