WebtoonMe: A Data-Centric Approach for Full-Body Portrait Stylization

TL;DR

This paper introduces a data-centric approach for full-body portrait stylization that overcomes previous limitations by enhancing dataset preparation, resulting in more plausible and diverse stylized outputs without changing the model architecture.

Contribution

The authors develop an advanced dataset preparation paradigm within a two-stage framework, enabling high-quality, diverse full-body portrait stylization in a production-level system.

Findings

Achieves high-quality stylization without additional losses

Produces more plausible and diverse full-body outputs

Improves robustness across skin tones

Abstract

Full-body portrait stylization, which aims to translate portrait photography into a cartoon style, has drawn attention recently. However, most methods have focused only on converting face regions, restraining the feasibility of use in real-world applications. A recently proposed two-stage method expands the rendering area to full bodies, but the outputs are less plausible and fail to achieve quality robustness of non-face regions. Furthermore, they cannot reflect diverse skin tones. In this study, we propose a data-centric solution to build a production-level full-body portrait stylization system. Based on the two-stage scheme, we construct a novel and advanced dataset preparation paradigm that can effectively resolve the aforementioned problems. Experiments reveal that with our pipeline, high-quality portrait stylization can be achieved without additional losses or architectural changes.