NPRportrait 1.0: A three-level benchmark for non-photorealistic rendering of portraits

Abstract Recently, there has been an upsurge of activity in image-based non-photorealistic rendering (NPR), and particular portrait image stylisation, due to the advent neural style transfer (NST). However, state performance evaluation this field is poor, especially compared norms computer vision machine learning communities. Unfortunately, task evaluating stylisation thus far not well defined, since it involves subjective, perceptual, aesthetic aspects. To make progress towards a solution, paper proposes new structured, three-level, benchmark dataset for stylised images. Rigorous criteria were used its construction, consistency was validated by user studies. Moreover, methodology developed algorithms, which makes use different levels as annotations provided studies regarding characteristics faces. We perform wide variety methods (both portrait-specific general purpose, also both traditional NPR approaches NST) using dataset.