Real Time Dress Code Adherence Recognition in an Academic Setting Using a Deep Learning Model

Abstract

Abstract: Machine learning is finding application in many fields as a tool. Its increasing adoption fueled by rapid advancements in algorithms and hardware. Deep learning techniques have shown great capabilities in image recognition, face recognition and other vision tasks. The proposed model describes the use of a deep learning method for the soft biometrics’ classification of clothing according to a predefined dress code standard in an academic setting. The Yolov4 architecture is used in this work for detection and classification. A custom dataset of images is gathered at a higher institution of learning by volunteer students which are subsequently box annotated for location of clothed figures. These are used for training and testing of the dress code detection model. The proposed model indicates detection by drawing bounding boxes and classifies by gender into appropriately dressed APD and not appropriately dressed NAPD. The results indicate that the proposed deep learning model is an efficient and successful network configuration for dress code detection and classification.