GenderVL-Bench

GenderVL-Bench is a compact vision-language benchmark for evaluating how Vision-Language Models (VLMs) interpret gender-related representations across different occupations.

Dataset

• 108 images
• 12 occupations
• 9 images per occupation
• Format: JPEG / ImageFolder
• Split: train

Usage

from datasets import load_dataset

dataset = load_dataset("suparnojit/GenderVL-Bench")

Citation

@dataset{sarkar2026gendervlbench,
  author    = {Sarkar, Suparnojit},
  title     = {GenderVL-Bench},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/suparnojit/GenderVL-Bench}
}