Published August 28, 2026 | Version v1.0.0
Dataset Open

Real-vs-AI face benchmark for vision-language models - data & code release

Authors/Creators

  • 1. Hwasung Medi-Science University
  • 2. Yonsei University
  • 3. Hoseo University
  • 4. Gwangju Institute of Science and Technology (GIST)

Description

Aggregate results, per-item model judgments and analysis code for a benchmark of 19 vision-language models judging real photographs versus ChatGPT-4o and Imagen 3 portrait recreations, under the same protocol as an earlier human study of 1,667 adults (arXiv:2603.24048). Data are CC BY 4.0; code is MIT (see LICENSE.md). The stimulus-code-to-filename mapping, collection seeds and verbatim model rationales are withheld; see README.md.

Files

gdrpaul3-byte/hsmu_vlm_detection_benchmark-v1.0.0.zip

Files (255.9 kB)

Name Size
md5:55b4d834daeb91341cfe1277aa5df7df
255.9 kB Preview Download

Additional details

Related works

Is supplement to
Software: https://github.com/gdrpaul3-byte/hsmu_vlm_detection_benchmark/tree/v1.0.0 (URL)

Software

Repository URL
https://github.com/gdrpaul3-byte/hsmu_vlm_detection_benchmark