Learning From Category-attribute Label for Image Attribute Recognition

In the field of image recognition, semantic attribute feature information of images plays a crucial role in enabling deep learning models to extract more comprehensive insights from the images. This allows the model to perform tasks such as fine-grained classification, zero-shot learning, and even model interpretability. However, compared to classification tasks that only require a single label, obtaining multiple semantic labels for images is often challenging. In the framework of weakly-supervised learning, partial-label learning allows models to learn a true label from multiple candidate labels. Relying on this learning paradigm, we propose Category-Attribute Labels Learning (CAL) based on the partial labels learning problem and corresponding learning algorithm to solve CAL problem. Category-attribute labels can be directly derived from textual descriptions of a category's attributes, which enable models to learn the true semantic features of images from low-cost attribute labels. In this paper, the category-attribute label problem has been formalized and shows through experiments that the model can effectively learn using category-attribute-labels. Our approach outperforms other state-of-the-art partial-label learning methods in terms of performance.