PaliGemma 1
Google DeepMind — Gemma
partial detail 15 method fields published
Evaluation
CountBenchQA
VLM-ready version of the CountBench dataset introduced because TallyQA was found lacking in its ability to assess current VLM counting ability (skewed number distribution and varying image quality)
How it was run
Open a release to see the setup, scoring, and other details that the lab published. Only fields the lab actually disclosed are shown.