Facial-sketch Synthesis: A New Challenge

Abstract This paper aims to conduct a comprehensive study on facial-sketch synthesis (FSS). However, due the high cost of obtaining hand-drawn sketch datasets, there is lack complete benchmark for assessing development FSS algorithms over last decade. We first introduce high-quality dataset FSS, named FS2K, which consists 2 104 image-sketch pairs spanning three types styles, image backgrounds, lighting conditions, skin colors, and facial attributes. FS2K differs from previous datasets in difficulty, diversity, scalability should thus facilitate progress research. Second, we present largest-scale investigation by reviewing 89 classic methods, including 25 handcrafted feature-based approaches, 29 general translation 35 image-to-sketch approaches. In addition, elaborate experiments existing 19 cutting-edge models. Third, simple baseline FSGAN. With only two straightforward components, i.e., facial-aware masking style-vector expansion, our FSGAN surpasses performance all state-of-the-art models proposed large margin. Finally, conclude with lessons learned past years point out several unsolved challenges. Our code available at https://github.com/DengPingFan/FSGAN .