Contrastive Visual Data Augmentation

Yu Zhou, Bingxuan Li, Mohan Tang +6

Large multimodal models (LMMs) often struggle to recognize novel concepts, as they rely on pre-trained knowledge and have limited ability to capture subtle visual details. Domain-s…