A 3B VLM trained with GRPO to call a zoom tool improves V*Bench accuracy by 5.7% over its base model but degrades TextVQA and HR-Bench performance.
Visual programming: Compositional visual reasoning without training
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
support 1representative citing papers
citing papers explorer
-
Reinforcing VLMs to Use Tools for Detailed Visual Reasoning Under Resource Constraints
A 3B VLM trained with GRPO to call a zoom tool improves V*Bench accuracy by 5.7% over its base model but degrades TextVQA and HR-Bench performance.