---
license: mit
task_categories:
- text-to-image
- image-to-image
language:
- en
tags:
- cultural-bias
- image-generation
- evaluation
- multi-cultural
- generative-ai
size_categories:
- 1K<n<10K
configs:
- config_name: add_attribution
  data_files: "add_attribution/*.csv"
- config_name: country_swap  
  data_files: "country_swap/*.csv"
- config_name: multi_loop_edits
  data_files: "multi_loop_edits/*/*.csv"
---

# ECB Datasets: Cultural Bias Evaluation in Generative Image Models

## Overview
This dataset contains human evaluation data for cultural bias analysis in image generation models, supporting the research paper "Exposing Blindspots: Cultural Bias Evaluation in Generative Image Models". ECB stands for "Evaluation Cultural Bias". The dataset includes prompts, generated images, and evaluation metrics across different countries and cultural contexts.

## Dataset Structure

### 1. Add Attribution (`add_attribution/`)
Contains step-by-step cultural attribution tasks where models are asked to modify images to represent specific cultural elements.

**Files:**
- `prompts.csv`: Contains prompts for cultural attribution tasks across different countries (Korea, China, India, Kenya, Nigeria, United States)

**Base Image:**
- The base image used for attribution tasks is available at: https://ecb-pub.s3.us-east-2.amazonaws.com/add_attribution/base.png

**Columns in prompts.csv:**
- `country`: Target country for cultural attribution
- `step`: Sequential step number (1-5)
- `prompt`: Specific instruction for that step
- `flux_output_file`: URL to Flux model output image
- `qwen_output_file`: URL to Qwen model output image

### 2. Country Swap (`country_swap/`)
Contains cross-cultural transformation tasks where images are transformed from one cultural style to another.

**Files:**
- `prompts.csv`: Contains prompts for country-to-country style transformations

**Columns in prompts.csv:**
- `input_file`: Source image URL
- `source_country`: Original country/culture
- `target_country`: Target country/culture for transformation
- `prompt`: Transformation instruction
- `output_path`: URL to transformed image

### 3. Model-Specific Datasets
Each model directory (`flux/`, `hidream/`, `nextstep/`, `qwen/`, `sd35/`) contains comprehensive evaluation data.

**Files:**
- `prompt-img-path_s3.csv`: Complete dataset with prompts and image paths for each model

**Columns in prompt-img-path_s3.csv:**
- `model`: AI model name
- `country`: Target country
- `category`: Image category (architecture, art, clothing, food, etc.)
- `sub_category`: Sub-category within main category
- `variant`: Style variant (general, modern, traditional)
- `T2I prompt`: Text-to-image generation prompt
- `I2I prompt`: Image-to-image editing prompt
- `base`: Base image URL
- `edit_1` to `edit_5`: Sequential edit image URLs

## Usage

This dataset is designed for:
- Cultural bias analysis in AI image generation
- Cross-cultural model evaluation
- Human-AI agreement studies
- Cultural representation research

### Loading the Dataset

```python
from datasets import load_dataset

# Load specific dataset configurations
add_attribution = load_dataset("seochan99/ecb-datasets", "add_attribution")
country_swap = load_dataset("seochan99/ecb-datasets", "country_swap")
multi_loop_edits = load_dataset("seochan99/ecb-datasets", "multi_loop_edits")

# Access data
print(add_attribution["train"][0])  # First example from add_attribution
print(country_swap["train"][0])    # First example from country_swap
print(multi_loop_edits["train"][0]) # First example from multi_loop_edits
```

## Authors

**Equal Contribution:** Huichan Seo, Sieun Choi, Minki Hong

**Authors:**
- Huichan Seo (Carnegie Mellon University)
- Sieun Choi (Dongguk University) 
- Minki Hong (Dongguk University)
- Yi Zhou (Carnegie Mellon University)
- Junseo Kim (Delft University of Technology)
- Lukman Ismaila (Johns Hopkins University)
- Naome Etori (University of Minnesota--Twin Cities)
- Mehul Agarwal (Carnegie Mellon University)
- Zhixuan Liu (Carnegie Mellon University)
- Jihie Kim (Dongguk University)
- Jean Oh (Carnegie Mellon University)

**Contact:** chans@andrew.cmu.edu

## Citation

If you use this dataset, please cite:
```
@dataset{ecb_datasets_2025,
  title={ECB Datasets: Cultural Bias Evaluation in Generative Image Models},
  author={Huichan Seo and Sieun Choi and Minki Hong and Yi Zhou and Junseo Kim and Lukman Ismaila and Naome Etori and Mehul Agarwal and Zhixuan Liu and Jihie Kim and Jean Oh},
  year={2025},
  url={https://huggingface.co/datasets/seochan99/ecb-datasets}
}
```

## Related Paper

This dataset supports the research paper:
```
@article{exposing_blindspots_2025,
  title={Exposing Blindspots: Cultural Bias Evaluation in Generative Image Models},
  author={Huichan Seo and Sieun Choi and Minki Hong and Yi Zhou and Junseo Kim and Lukman Ismaila and Naome Etori and Mehul Agarwal and Zhixuan Liu and Jihie Kim and Jean Oh},
  year={2025},
  journal={[Journal Name]}
}
```

## License

This dataset is released under the **MIT License** for research and educational purposes.

**MIT License**
```
MIT License

Copyright (c) 2025 ECB Research Team

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
```

**Usage Terms:**
- ✅ Research and educational use
- ✅ Commercial use allowed
- ✅ Modification and distribution allowed
- ✅ Attribution required
- ❌ No warranty provided

## Contact

chans@andrew.cmu.edu