ShyFoo/CountHallu-dataset-RealHand

CountHalluSet — RealHand

Real-image dataset from Counting Hallucinations in Diffusion Models (arXiv:2510.13080). Part of CountHalluSet, a suite with well-defined counting criteria used to measure counting hallucination. Here the counting object is the five fingers of a human hand — a hand with the wrong number of fingers is a hallucination.

What's inside

Photos of single human hands, each paired with a segmentation mask sharing the same file stem. The image + mask pairing enables joint image+mask diffusion (JDM). Images are used at 256×256 during training.

RealHand/
├── images/   # hand photos
└── masks/    # matching segmentation masks (same stem as the image)

There is no labels.csv: finger counts are not stored, they are measured at evaluation time by the RealHand counter (CountHallu-counting_model-RealHand), gated by the quality classifier (CountHallu-quality_cls_model-RealHand).

Source & license

Usage

huggingface-cli download ShyFoo/CountHallu-dataset-RealHand \
    --repo-type dataset --local-dir $DATASET_ROOT/RealHand

Load with the reference code (counthallu.datasets.RealHand). See the CountHallu repository for the two-stage SD-1.5 JDM recipe and the full evaluation protocol.

Citation

@article{fu2025counting,
  title={Counting Hallucinations in Diffusion Models},
  author={Fu, Shuai and Zhou, Jian and Chen, Qi and Jing, Huang and Nguyen, Huy Anh and Liu, Xiaohan and Zeng, Zhixiong and Ma, Lin and Zhang, Quanshi and Wu, Qi},
  journal={arXiv preprint arXiv:2510.13080},
  year={2025}
}