State-of-the-art vision-language models score only about 20% on a new synthetic color vision test, and LoRA fine-tuning on the same test distribution raises accuracy to about 94%.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Assessing Color Vision Test in Large Vision-language Models
State-of-the-art vision-language models score only about 20% on a new synthetic color vision test, and LoRA fine-tuning on the same test distribution raises accuracy to about 94%.