UniFGVC: Universal Training-Free Few-Shot Fine-Grained Vision Classification via Attribute-Aware Multimodal Retrieval
UniFGVC is a training-free method for fine-grained visual classification that treats it as a multimodal retrieval problem. It uses a Category-Discriminative Visual Captioner (CDV-Captioner) to create structured, attribute-rich image descriptions from few samples. This allows the system to recognize new categories by finding the closest matching templates in a database, bypassing the need for parameter fine-tuning and mitigating overfitting.