Automated Fashion Clothing Image Labeling System

As interest in fashion clothing has recently increased, the importance of fashion services using deep learning technology is increasing. These services are used in a variety of fields, including customized clothing recommendations, fashion style analysis, and improving online shopping experiences. However, the image labeling work, which is the core of these services, is still performed manually, which requires high costs and time, and relies on manpower. This study seeks to automate this process of fashion clothing labeling by using three steps image preprocessing technology and Yolov8 for dataset learning. During the research process, rembg and OpenCV were used to preprocess image data, and OpenPose was introduced to obtain object joint information for the accurate location of clothing items. Through this, we were able to effectively remove the background of the image, extract and classify the joints of the object, distinguish tops and bottoms, and then teach them to Yolo. As a result, this study presents an automated labeling approach that can maintain high accuracy while reducing the time and cost required for manual labeling of fashion clothing images.