Larysa Koriashkina

Kostiantyn Khabarlak, Larysa Koriashkina

Dense facial landmark detection is one of the key elements of face processing pipeline. It is used in virtual face reenactment, emotion recognition, driver status tracking, etc. Early approaches were suitable for facial landmark detection in controlled environments only, which is clearly insufficient. Neural networks have shown an astonishing qualitative improvement for in-the-wild face landmark detection problem, and are now being studied by many researchers in the field. Numerous bright ideas are proposed, often complimentary to each other. However, exploration of the whole volume of novel approaches is quite challenging. Therefore, we present this survey, where we summarize state-of-the-art algorithms into categories, provide a comparison of recently introduced in-the-wild datasets (e.g., 300W, AFLW, COFW, WFLW) that contain images with large pose, face occlusion, taken in unconstrained conditions. In addition to quality, applications require fast inference, and preferably on mobile devices. Hence, we include information about algorithm inference speed both on desktop and mobile hardware, which is rarely studied. Importantly, we highlight problems of algorithms, their applications, vulnerabilities, and briefly touch on established methods. We hope that the reader will find many novel ideas, will see how the algorithms are used in applications, which will enable further research.

Kostiantyn Khabarlak, Larysa Koriashkina

Neural networks are now actively being used for computer vision tasks in security critical areas such as robotics, face recognition, autonomous vehicles yet their safety is under question after the discovery of adversarial attacks. In this paper we develop simplified adversarial attack algorithms based on a scoping idea, which enables execution of fast adversarial attacks that minimize structural image quality (SSIM) loss, allows performing efficient transfer attacks with low target inference network call count and opens a possibility of an attack using pen-only drawings on a paper for the MNIST handwritten digit dataset. The presented adversarial attack analysis and the idea of attack scoping can be easily expanded to different datasets, thus making the paper's results applicable to a wide range of practical tasks.