Generated Faces in the Wild (GFW) is a comprehensive dataset used to conduct a quantitative comparison of faces generated by multiple image generation models, including Stable Diffusion, Midjourney, and DALL-E. This dataset contains a large number of generated images alongside annotated bounding boxes that define the locations of faces within the images.
(Arxiv)
The goal of the dataset is to provide an in-depth analysis of the quality, diversity, and accuracy of faces generated by these popular AI-based image generation models.
This dataset compares and evaluates the performance of several image generation models, namely:
- Stable Diffusion
- Midjourney
- DALL-E
The GFW dataset includes:
- Generated Images: A large collection of images generated by Stable Diffusion, Midjourney, and DALL-E.
- Annotated Faces: For each image, bounding boxes are provided to annotate the locations of faces within the images.
The dataset serves as a valuable resource for comparing and analyzing:
- The quality of generated faces by different models.
- The ability of models to accurately place faces within the generated images.
- The diversity of faces generated by each model across various categories.
Researchers and practitioners can use this dataset to evaluate how these models perform in terms of realistic face generation, facial diversity, and alignment with typical human characteristics.
- Multiple Models Compared: The dataset includes images generated by Stable Diffusion, Midjourney, and DALL-E. Each model's performance is compared in terms of the quality and diversity of faces generated.
- Bounding Box Annotations: Each generated face is annotated with bounding boxes, providing accurate coordinates of face locations in the images. This makes it possible to perform further analysis on the placement, alignment, and proportion of faces.
- Large Dataset: The dataset provides a substantial number of images for comprehensive comparisons and evaluations. The large number of faces allows for a robust quantitative analysis.
The dataset is available for download at the following link:
You can access and explore the dataset by visiting the link, which contains both the images and the corresponding annotations.
- Number of Images: X images across multiple models (Stable Diffusion, Midjourney, DALL-E).
- Annotations: Bounding boxes for each face in the images, stored in a structured format (e.g., JSON, CSV).
- Models Compared: Stable Diffusion, Midjourney, DALL-E (and possibly others in future versions).
This dataset can be utilized for various purposes:
- Face Detection Research: Researchers working on face detection and recognition can use this dataset to test and benchmark face detection algorithms on AI-generated images.
- Model Performance Evaluation: The dataset is valuable for comparing the performance of different image generation models, especially in terms of realism and accuracy of face generation.
- Face Generation Quality Analysis: The dataset can be used to evaluate how different generative models produce faces with respect to alignment, diversity, and other facial features.
If you use the dataset for your research, please cite the original paper:
@inproceedings{borjiGFW,
title={Generated Faces in the Wild: Quantitative Comparison of Stable Diffusion, Midjourney and DALL-E 2},
author={Ali Borji},
booktitle={Arxiv},
year={2022},
url={https://arxiv.org/pdf/2210.00586.pdf}
}
This dataset is made available under the [INSERT LICENSE TYPE HERE] license. Please refer to the LICENSE file for more details.
If you would like to contribute to this project, feel free to fork the repository and submit pull requests with improvements or additions to the dataset.