Cross-modal Collaboration for Augmented Few-Shot Image Classification

Training high-performance image classification models typically requires a large amount of labeled data. However, in practical applications, obtaining sufficient and high-quality training data is often challenging due to high data collection costs and annotation difficulties. To address this, we propose the Text and Image Collaboration-based Augmented Few-Shot Learning method (TICA-FSL). TICA-FSL leverages auxiliary modality data to generate diverse sample features through collaborative augmentation. By incorporating text features into the visual embedding space, it enhances the performance and robustness of classification models under data-scarce conditions. Experiments on CUB-200 and Oxford-102 datasets show that TICA-FSL significantly outperforms baseline methods with accuracy improvements of 4%and 2%respectively in 5-shot and 1-shot scenarios. The results demonstrate the effectiveness of utilizing multi-modal data from text and images for few-shot learning, expanding the capabilities of deep learning in practical applications.