Google DeepMind introduced PaliGemma 2, a new family of Vision-Language Models with parameter sizes ranging from 3 billion to 28 billion, designed to address challenges in generalizing across different tasks and adapting to various input data types, including diverse image resolutions.
