Skip to content

Anh Totti Nguyen

Associate Professor of Computer Science, Auburn University

  • Current Page Parent Research
  • Lab
  • Press
  • Work with me
  • Teaching
    • Courses
    • K-6 AI club
  • About
  • CV
  • Current Page Parent Research
  • Lab
  • Press
  • Work with me
  • Teaching
    • Courses
    • K-6 AI club
  • About
  • CV

vlms_biased_accuracy

May 2025 0

All VLMs achieve 100% on identification and counting tasks with unmodified images, showing that they fully recognize the original version but fail on the counting questions on the modified images (i.e., counterfactuals) in VLMBias. The mean accuracy of five state-of-the-art VLMs on our seven tasks is 17.05%.

Share
  • Previous Vision Language Models are Biased
Computer Vision

ImageNet-Hard: The Hardest Images Remaining from a Study of the Power of Zoom and Spatial Biases in Image Classification

  • June 10, 2023
Computer Vision

Vision Language Models Are Blind

  • July 10, 2024
A JavaScript Module Pattern
Great Articles Javascript

A JavaScript Module Pattern

  • December 28, 2009
  • 1
Computer Vision

GlitchBench: Can large multimodal models detect video game glitches?

  • April 10, 2024

Anh Totti Nguyen © 2026. All Rights Reserved.

Powered by WordPress. Theme by Alx.