Watch Classification Using DL Models

This paper presents a comprehensive pipeline for detecting and classifying Rado watch models in high-quality images using state-of-the-art deep learning techniques. The proposed approach sequentially integrates four main stages: YOLOv5 for initial object detection, the Segment Anything Model (SAM) for precise segmentation of the detected watches, Real-ESRGAN to enhance image quality through super-resolution on the segmented watch regions, and Vision Transformer (ViT) for final model classification. This pipeline is specifically designed to address the challenges in luxury product identification, where achieving high accuracy and preserving intricate details are essential. Experimental results on a diverse dataset of watch images indicate that this approach achieves a mean Average Precision (mAP) of 95% in detection and a classification accuracy of 93%, surpassing traditional methodologies. The results underscore the effectiveness of this integrated approach, combining cutting-edge detection, segmentation, enhancement, and classification techniques to deliver a reliable solution for applications in luxury product classification and digital marketing.