Datasets:
license: apache-2.0
pretty_name: CountHalluSet — ToyShape
task_categories:
- unconditional-image-generation
- image-classification
size_categories:
- 10K<n<100K
tags:
- diffusion
- counting
- hallucination
- synthetic
CountHalluSet — ToyShape
Synthetic 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 — a diffusion model generating the wrong number of instances, even for patterns absent from its training data.
What's inside
128×128 RGB images of non-overlapping white shapes on a black background. Each
image holds 1–3 shapes drawn from {triangle, square, pentagon}, at most one
instance per type. This defines the counting criterion: a correct sample has
each present shape exactly once; two of any shape, or an empty image, is a
hallucination.
ToyShape/
├── images/ # 00000.png, 00001.png, ...
└── labels.csv # filename, triangle, square, pentagon (each count ∈ {0, 1})
Default release: 30,000 samples.
Usage
huggingface-cli download ShyFoo/CountHallu-dataset-ToyShape \
--repo-type dataset --local-dir $DATASET_ROOT/ToyShape
Load with the reference code (counthallu.datasets.ToyShape) or regenerate from
scratch — the generator is deterministic given a seed:
python -m counthallu.datasets.toyshape --data_root $DATASET_ROOT --num_samples 30000
See the CountHallu repository for training 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}
}