Flickr-Faces-HQ Dataset with 70,000 High-Quality Images

The Flickr-Faces-HQ (FFHQ) dataset contains 70,000 high-quality PNG images of human faces. It was originally created by NVIDIA researchers as a benchmark for generative adversarial networks (GANs).

Use Cases

• Train or benchmark StyleGAN architectures on 70,000 high-quality PNG images of human faces.
• Develop face generation models using the high-quality PNG images as a standard training corpus.
• Conduct facial attribute analysis or editing tasks using the high-quality PNG images from the FFHQ collection.

Strengths

• Contains 70,000 images, providing a substantial volume of data for model training.
• Images are described as high-quality PNGs, suitable for detailed visual tasks.
• Serves as a known benchmark dataset for generative adversarial network research.

Limitations

• The specific attributes, labels, or metadata for the 70,000 images are not detailed, limiting supervised learning applications.
• The dataset's demographic or geographic distribution is unknown, which may introduce bias.
• The original source and curation method for the images are not fully described in the provided input.

Provenance

Source

Originally created by Tero Karras, Samuli Laine, and Timo Aila of NVIDIA.

Collection Method

Created as a benchmark dataset for GAN research; specific collection methodology not detailed.

Freshness

The dataset was last updated on the platform in April 2022.