A Comparative Analysis on Fashion Cloth Feature Extraction Methods using Text and Image Data

Modern methods of extracting relevant features from images and texts are crucial to the recovery of fashionable garments. In this research, we compare and contrast a number of different approaches to extracting features from images and text. We utilize Generative Pre-trained Transformer (GPT) and Bidirectional Encoder Representations from Transformers (BERT) to efficiently process and represent textual descriptions of fashion products for text feature extraction. This allows us to take advantage of GPT's generative capabilities and contextual understanding. In order to extract features from photographs of fashion sketch images, we use a mix of Residual Networks (ResNet) and Visual Geometry Group (VGG) networks. This allows us to pick up complex visual patterns and details. Our method is designed to enhance the accuracy of text and image matching by combining the best features of these state-of-the-art models. Accuracy, recall, F1-score, and retrieval efficiency are some of the assessment metrics that were examined on a wide dataset that included different types of fashion clothing. The hybrid model improves feature extraction capabilities, producing robust and discriminative text and picture representations, according to preliminary findings. More intelligent and user-friendly retrieval systems in digital fashion platforms and e-commerce might be possible thanks to the study's useful insights into the possibilities of advanced deep learning architectures in the field of fashion product retrieval.