Cloth Extraction and Restoration from DeepFashion for Pre-training CNNs in Laundry Classification

In the realm of computer vision and artificial intelligence, the classification of objects plays a significant role in various applications, like classifying laundry to assist sorting. The process of generating and annotating large volumes of images for training convolutional neural networks (CNNs) can be both time-consuming and costly. This paper presents a novel approach to enhance the training of CNNs for laundry classification by leveraging fashion images. Our methodology involves extracting clothing items from fashion images, detecting and restoring not visible parts, and utilizing the resulting dataset for pre-training CNNs. Through this process, the CNN learns to discern and classify various laundry items more effectively, benefiting from the rich diversity and complexity present in fashion imagery. We demonstrate the efficacy of our approach through extensive experimentation and evaluation on a benchmark dataset and using explainable AI, showcasing notable improvements in laundry classification accuracy compared to conventional training methods. The proposed framework not only advances the state-of-the-art in laundry classification but also underscores the potential of utilizing public imagery for enhancing CNN training in related domains.