500 Ishihara-like images testing VLM robustness on visually adversarial scenarios. Numbers 0-99 rendered in different fonts and color configurations against confusing backgrounds.
Field
Description
image
Ishihara-like test image
filename
Image filename
number
Hidden number in the image (ground truth, 0-99)
font
Font used (Arial / DejaVuSans)
color_config
Color configuration variant
prompt: "What… See the full description on the dataset page:
https://huggingface.co/datasets/MM-Hallu/ColorBlindnessEval.