Real-Time Portrait Stylization on the Edge

TL;DR

This paper presents a real-time portrait stylization method optimized for mobile devices, achieving high-quality cartoon and anime effects with significantly reduced computation.

Contribution

It introduces a latency-driven differentiable architecture search that reduces model complexity and enables real-time stylization on smartphones.

Findings

10x reduction in computation compared to baseline models

Real-time stylization achieved on off-the-shelf smartphones

Maintains high-quality, realistic stylized portraits

Abstract

In this work we demonstrate real-time portrait stylization, specifically, translating self-portrait into cartoon or anime style on mobile devices. We propose a latency-driven differentiable architecture search method, maintaining realistic generative quality. With our framework, we obtain $10\times$ computation reduction on the generative model and achieve real-time video stylization on off-the-shelf smartphone using mobile GPUs.