Clothing image classification plays a pivotal role in modern fashion e-commerce, aiding in inventory management, recommendation systems, and visual search. This project by Zhu Hongyun explores the use of ResNet, a deep residual learning framework, to classify clothing images into various categories. The model was trained on a curated dataset of five categories: bags, dress, pants, shorts, and upper wear. The ResNet architecture enables the network to maintain accuracy even as it deepens by employing identity shortcut connections to combat vanishing gradients. Key preprocessing steps included background removal to improve focus on clothing contours. Performance metrics such as accuracy, loss, and F1-score were used to validate the model. The final implementation integrates a user-friendly GUI for selecting an image and displaying classification results.
The model demonstrates significant potential for improving clothing categorization in fashion applications. The integration of deep residual learning with clean, preprocessed data enhances recognition capabilities across varied apparel types. The GUI application makes this technology accessible to non-experts, broadening its practical impact.
The model could be expanded to handle accessories like hats and shoes or integrated with multi-label classification to support outfits. Deployment to cloud platforms or integration with fashion e-commerce APIs could enable real-time usage at scale.