PersonXray_350k Dataset

Welcome to the PersonXray dataset, a comprehensive collection of high-quality portrait images designed for research and development in the fields of identity recognition, image generation, and multimodal learning.

Overview

The PersonXray dataset contains approximately 350,000 portrait images, each with a corresponding textual description. The images have been carefully processed to remove backgrounds, focusing solely on the subject's portrait. This dataset is ideal for training and testing models that require understanding and generation of human identity features.

Key Features

• Background Removed: Each image has a transparent background, highlighting the subject.
• Diverse Demographics: The dataset includes a wide range of ages, ethnicities, and expressions.
• Textual Descriptions: Each image is accompanied by a detailed description, enriching the multimodal learning experience.
• Open Source: Available under an open-source license for unrestricted use.

Usage

The PersonXray dataset can be used for various applications, including but not limited to:

• Training deep learning models for facial recognition and verification.
• Developing generative models that create new images with specific identity traits.
• Analyzing the relationship between visual and textual data for multimodal AI systems.

Data Structure

The dataset is organized as follows:

• source_images/ : Contains all the source images in JPEG format.

• processed_images/: Contains all the processed portrait images in JPEG format.

• descriptions/: Contains text files with descriptions corresponding to each image.

Each image and its description are named consistently to ease the pairing process.

Getting Started

To get started with the PersonXray_350k dataset, follow these steps:

1. Unzip: First, unzip the dataset file to extract the folders (source_images/, processed_images/, and descriptions/).
2. Access Data: Access the source_images/ ,processed_images/ and descriptions/ folders to begin using the data.
3. Browse Descriptions: Browse the descriptions/ folder to find caption_list.txt to determine which images have descriptions.

Prerequisites

Before using the dataset, ensure you have the necessary software and libraries installed, such as Python and relevant deep learning frameworks.

Installation

If you plan to use the dataset with a specific framework or library, follow their installation guidelines. Otherwise, the dataset can be used directly without additional installation.

License

This dataset is licensed under the Creative Commons Attribution 4.0 International License. Please refer to the license for details on how to properly use and attribute the dataset.

Contributing

We welcome contributions to the PersonXray dataset. If you have additional images or descriptions to add, please submit a pull request or contact the maintainers.

Contact

For questions or further information, please contact Weijue Bu at weijuebu@gmail.com.

Citation

If you use the PersonXray dataset in your research, please cite it as:

Weijue Bu, et al. "PersonXray_350k: A High-Quality Portrait and Description Dataset." Hugging Face, https://huggingface.co/datasets/MuQYY/PersonXray_350k, 2024.

---