Benchmarking Shadow Removal for Facial Landmark Detection

Facial landmark detection is a very fundamental task and its accuracy plays a significant role for many downstream face-related vision applications. In practice, the facial landmark detection can be affected by a lot of natural degradations. One of the most common and important degradations is the shadow caused by light source being blocked with an external occluder. While many advanced shadow removal methods have been proposed to restore the image quality in recent years, their effects on facial landmark detection are not well studied. For example, it remains unclear whether the shadow removal could enhance the robustness of facial landmark detection to diverse shadow patterns or not. In this work, for the first time, we construct a novel benchmark (i.e., SHAREL) to link the two independent but relatable tasks (i.e., shadow removal and facial landmark detection). In particular, SHAREL covers diverse face shadows with different intensities, sizes, shapes, and locations. Moreover, to mine hard shadow patterns against facial landmark detection, we propose a novel method (i.e., adversarial shadow attack), which allows us to construct a challenging subset of the benchmark for a comprehensive analysis. With the constructed benchmark, we conduct extensive analysis on three state-of-the-art shadow removal methods and three landmark detectors. We observed a highly positive correlation between shadow removal and facial landmark detection tasks, which probably will provide insight to improve the robustness of the facial landmark detection in the future.