README.md

metadata

license: mit
language:
  - en
tags:
  - text-image
  - adversarial
  - real-world
  - unusual
  - computer-vision
  - multimodal
  - image
  - text
size_categories:
  - n<1K

(Un-)usual objects in (un-)usual places

A small Text-Image dataset to confuse, probe (and improve) SOTA (2024) machine vision models.

To be continued (with further examples added)...

Example results from LMSYS ARENA (June 2024):