---
license: mit
task_categories:
- image-classification
- visual-question-answering
language:
- en
tags:
- visual-illusions
- pareidolia
- multimodal
- synthetic
- fashion-mnist
pretty_name: IllusionFashionMNIST Training Set
size_categories:
- 1K<n<10K
---

# IllusionFashionMNIST — Training Set

## Dataset summary

This repository contains the training split of **IllusionFashionMNIST**, one of the four datasets introduced in *Illusory VQA: Benchmarking and Enhancing Multimodal Models on Visual Illusions*. It is designed to train and evaluate models on the recognition of Fashion-MNIST categories embedded as visual illusions (pareidolia) in generated scenes.

The source-condition images are sampled from Fashion-MNIST and resized to 512 × 512 pixels. They are combined with English scene descriptions and transformed into illusory images using ControlNet. The split also includes a balanced **No illusion** class.

| Property | Value |
| --- | --- |
| Hugging Face repository | [VQA-Illusion/FashionMnist_train](https://huggingface.co/datasets/VQA-Illusion/FashionMnist_train) |
| Official split | Train |
| Task | 11-way illusion classification / visual question answering |
| Annotated examples | 3,300 |
| Image format | JPEG |
| Metadata file | <code>df_data.csv</code> |
| Paper | [arXiv:2412.08169](https://arxiv.org/abs/2412.08169) |
| Code | [IllusoryVQA/IllusoryVQA](https://github.com/IllusoryVQA/IllusoryVQA) |

## Repository structure

| Path | Files | Description |
| --- | ---: | --- |
| <code>ill_images/</code> | 3,300 | Primary training images. The folder contains both illusion-bearing examples and the 300 examples assigned to the No illusion class. |
| <code>raw_images/</code> | 3,000 | Source-condition images used to generate the illusion-bearing examples. No raw counterpart is provided for the 300 No illusion rows. |
| <code>df_data.csv</code> | 1 | Canonical metadata and target labels. |
| <code>README.md</code> | 1 | Dataset card. |

The value in <code>image_name</code> is the filename stem. For example, <code>FashionMnist_train_0</code> corresponds to <code>ill_images/FashionMnist_train_0.jpg</code>.

## Metadata schema

The CSV contains one row per primary training example.

| Column | Type | Description |
| --- | --- | --- |
| <code>image_name</code> | string | Image identifier and filename stem. |
| <code>Pprompt</code> | string | Positive scene prompt used during illusion-image generation. |
| <code>Nprompt</code> | string | Negative generation prompt. It is empty for No illusion rows. |
| <code>illusion_strength</code> | float or empty | Control strength used for illusion generation. It is <code>1.5</code> for illusion-bearing rows and empty for No illusion rows. |
| <code>label</code> | string | Ground-truth Fashion-MNIST class ID (<code>0</code>–<code>9</code>) or <code>No Illusion</code>. |

Read the CSV with string-preserving options because <code>label</code> contains both numbers and text:

~~~python
import pandas as pd

metadata = pd.read_csv(
    "df_data.csv",
    dtype={"label": "string"},
    keep_default_na=False,
)
metadata["label_normalized"] = metadata["label"].str.strip().str.casefold()
~~~

## Label mapping

The following order matches Fashion-MNIST and the class lists used in the official experiment code.

| Numeric ID | Class label | Value stored in <code>df_data.csv</code> |
| ---: | --- | --- |
| 0 | T-shirt/top | <code>0</code> |
| 1 | Trouser | <code>1</code> |
| 2 | Pullover | <code>2</code> |
| 3 | Dress | <code>3</code> |
| 4 | Coat | <code>4</code> |
| 5 | Sandal | <code>5</code> |
| 6 | Shirt | <code>6</code> |
| 7 | Sneaker | <code>7</code> |
| 8 | Bag | <code>8</code> |
| 9 | Ankle boot | <code>9</code> |
| 10 | No illusion | <code>No Illusion</code> |

Use ID 10 when converting the textual No illusion target to a fully numeric label space.

## Download

Install the required packages:

~~~bash
pip install -U huggingface_hub pandas pillow
~~~

Download the complete repository:

~~~python
from huggingface_hub import snapshot_download

dataset_dir = snapshot_download(
    repo_id="VQA-Illusion/FashionMnist_train",
    repo_type="dataset",
)
print(dataset_dir)
~~~

Command-line alternative:

~~~bash
huggingface-cli download VQA-Illusion/FashionMnist_train \
  --repo-type dataset \
  --local-dir FashionMnist_train
~~~

## Load and pair images with metadata

Use <code>df_data.csv</code> as the source of truth rather than treating the top-level image directories as semantic classes.

~~~python
from pathlib import Path
import pandas as pd
from huggingface_hub import snapshot_download

root = Path(snapshot_download(
    repo_id="VQA-Illusion/FashionMnist_train",
    repo_type="dataset",
))

df = pd.read_csv(
    root / "df_data.csv",
    dtype={"label": "string"},
    keep_default_na=False,
)

df["illusion_path"] = df["image_name"].map(
    lambda name: root / "ill_images" / (name + ".jpg")
)
df["raw_path"] = df["image_name"].map(
    lambda name: root / "raw_images" / (name + ".jpg")
)
df["raw_path"] = df["raw_path"].map(
    lambda path: path if path.exists() else None
)

id_to_label = {
    0: "T-shirt/top",
    1: "Trouser",
    2: "Pullover",
    3: "Dress",
    4: "Coat",
    5: "Sandal",
    6: "Shirt",
    7: "Sneaker",
    8: "Bag",
    9: "Ankle boot",
    10: "No illusion",
}

def to_label_id(value):
    value = str(value).strip()
    return 10 if value.casefold() == "no illusion" else int(value)

df["label_id"] = df["label"].map(to_label_id)
assert df["illusion_path"].map(Path.exists).all()
~~~

## Intended use

Suitable uses include:

- supervised training for illusion-aware image classification;
- VQA-style prompting over a fixed answer vocabulary;
- paired comparisons between source-condition and generated illusion images; and
- robustness analysis for multimodal and vision-language models.

For VQA evaluation, ask whether the image contains an illusion and restrict the answer space to the ten Fashion-MNIST classes plus No illusion. The exact prompt templates used in the paper are available in Appendix G and in the official code repository.

## Dataset creation and safety

The authors generated diverse English scene descriptions with several language models, combined those descriptions with source-condition images, and produced the illusion images with ControlNet. Human reviewers validated dataset quality. The paper also reports content-safety screening and the removal of images flagged by NSFW detectors from the public release.

## Important usage notes

- <code>df_data.csv</code> is the authoritative index. Do not infer the semantic target from the top-level directory name.
- Hugging Face may automatically display the top-level image folders as <code>imagefolder</code> classes. Those folder-derived labels describe image variants, not Fashion-MNIST targets.
- The No illusion label is textual while the other labels are numeric strings; normalize it explicitly as shown above.
- Only illusion-bearing rows have files in <code>raw_images/</code>.
- This benchmark primarily covers a single large hidden category per image. See the paper for the full limitations.

## License

This dataset repository declares the MIT license. Users should also review and comply with any applicable terms associated with Fashion-MNIST and other upstream components.

## Citation

If you use this dataset, please cite:

~~~bibtex
@misc{rostamkhani2024illusoryvqa,
  title        = {Illusory VQA: Benchmarking and Enhancing Multimodal Models on Visual Illusions},
  author       = {Rostamkhani, Mohammadmostafa and Ansari, Baktash and Sabzevari, Hoorieh and Rahmani, Farzan and Eetemadi, Sauleh},
  year         = {2024},
  eprint       = {2412.08169},
  archivePrefix = {arXiv},
  primaryClass = {cs.CV},
  url          = {https://arxiv.org/abs/2412.08169}
}
~~~

## Contact

Questions and reproducibility issues can be submitted through the [official GitHub repository](https://github.com/IllusoryVQA/IllusoryVQA).