Towards Robust Object Detection: Bayesian RetinaNet for Homoscedastic Aleatoric Uncertainty Modeling

According to recent studies, commonly used computer vision datasets contain\nabout 4% of label errors. For example, the COCO dataset is known for its high\nlevel of noise in data labels, which limits its use for training robust neural\ndeep architectures in a real-world scenario. To model such a noise, in this\npaper we have proposed the homoscedastic aleatoric uncertainty estimation, and\npresent a series of novel loss functions to address the problem of image object\ndetection at scale. Specifically, the proposed functions are based on Bayesian\ninference and we have incorporated them into the common community-adopted\nobject detection deep learning architecture RetinaNet. We have also shown that\nmodeling of homoscedastic aleatoric uncertainty using our novel functions\nallows to increase the model interpretability and to improve the object\ndetection performance being evaluated on the COCO dataset.\n