ArtFID: Quantitative Evaluation of Neural Style Transfer

The field of neural style transfer has experienced a surge research exploring different avenues ranging from optimization-based approaches and feed-forward models to meta-learning methods. developed techniques have not just progressed the transfer, but also led breakthroughs in other areas computer vision, such as all visual synthesis. However, whereas quantitative evaluation benchmarking become pillars vision research, reproducible, assessment is still lacking. Even comparison fields synthesis, where widely used metrics exist, lagging behind. To support automatic study their respective strengths weaknesses, would greatly benefit measurement stylization performance. Therefore, we propose method complement currently mostly qualitative schemes. We provide extensive evaluations large-scale user show that proposed metric strongly coincides with human judgment.