A dataset comprising 54,210 groups of synthetic images, flaws (14 categories) and corresponding explanations. The ratings of relevant phrases across 3 refinement iterations are also included.
Paper: [AAAI-26] Explainable Synthetic Image Detection through Diffusion Timestep Ensembling
Code: https://github.com/Shadowlized/ESIDE
Dataset Example
{
"flaw_id": 1,
"flaw_category": "Lighting",
"img_path":… See the full description on the dataset page: https://huggingface.co/datasets/Shadowlized/GenExplain.
