Clothing Detection and Classification with Fine-Tuned YOLO-Based Models

Clothing helps to enhance the wearer's appearance, so many people pay attention to choosing clothes to suit the events they attend. Nowadays, e-commerce websites are growing strongly, while products are also increasingly diverse and abundant in all areas of life, especially apparel, and clothes with many different types and brands. Therefore, finding appropriate clothes with only an image is challenging. This collected ordinary clothes types in Vietnam, then fine-tuned the model of You Only Look Once (YOLO) version of 5l to provide better accuracy with smaller image size in clothes recognition in the image compared to the original one YOLOv5l with default values of hyper-parameters, and previous YOLO versions. As a result, the method obtains an accuracy of 0.933 in clothes detection on over 10,000 images of 18 clothes types, including popular clothes in Vietnam. The work is expected to provide an image-based useful search for e-commerce systems.