What is it about?

The paper introduces an automatic architecture for portrait stylization. Given limited style exemplars (∼100), the new architecture can produce high-quality style transfer results with advanced ability to synthesize high-fidelity contents, strong generality to handle complicated scenes (e.g., occlusions and accessories), and high scalability to full-body translation with only head observations.

Featured Image

Why is it important?

We define the new rule of "calibration first, translation later" for few-shot image translation task, and explore the augmented global structure with locally-focused translation. This elegant design makes all-around quality boosts compared with previous methods.


It was a great pleasure to write this article. I hope that this article will help people to create artistic contents more efficient and simpler.

Yifang Men
Alibaba Group

Read the Original

This page is a summary of: DCT-net, ACM Transactions on Graphics, July 2022, ACM (Association for Computing Machinery), DOI: 10.1145/3528223.3530159.
You can read the full text:



The following have contributed to this page