PaliGemma 2 was released last week! It’s the next evolution of the first vision-language model in the Gemma model family, available in multiple model sizes (3B, 10B, 28B parameters) and resolutions (224px, 448px, 896px).
I have a semantic segmentation dataset with labels in a list-of-polygon-points format. How should I format it for fine-tuning with either the PaliGemma or PeliGemma 2 model?