PaliGemma 2: A Family of Versatile VLMs for Transfer

Benchmark Model Rank Results
image-captioning-on-textcapsPaliGemma 2 10B#1CIDEr: 157.7
visual-grounding-on-refcoco-valPaliGemma 2 10B#7Accuracy: 78.2
visual-question-answering-vqa-on-textvqa-valPaliGemma 2 10B#18Accuracy: 76.6