INDEPENDENT MINDS. EXCEPTIONAL POSSIBILITIES.Independent minds. Considered guidance.
Apply UYOUR FUTURE, THOUGHTFULLY.中文Book a consultation

Research & ideas Event year 2025

ColorBlindnessEval accepted at the SCI-FM Workshop at ICLR 2025

The Apply U team’s ColorBlindnessEval paper was accepted at the Open Science for Foundation Models (SCI-FM) Workshop at ICLR 2025. The team updated its acceptance announcement on 6 March 2025, and the public preprint also records the workshop acceptance.

Edited by Apply U1 min read

At a glance

  • Accepted at the SCI-FM Workshop at ICLR 2025
  • A dataset of 500 Ishihara-like images
  • Evaluation of nine vision-language models

Research question and evaluation method

ColorBlindnessEval examines how vision-language models recognise numbers within complex colour patterns. The paper describes a dataset of 500 images resembling Ishihara colour-vision tests and uses it to evaluate nine models.[1]

Vision-language models process both images and text. The study uses number recognition as a specific task for examining model performance and reliability when interpreting colour patterns.[1]

Acceptance and public paper

The SCI-FM Workshop focuses on open science, reproducibility and foundation-model evaluation. Both the Apply U announcement and the arXiv record identify the acceptance as belonging to this workshop.[1][2][3]

The full paper title is ColorBlindnessEval: Can Vision-Language Models Pass Color Blindness Tests? The arXiv page lists Zijian Ling, Han Zhang, Yazhuo Zhou and Jiahao Cui as its authors.[1]

Back to the newsroom