21 Million Facial Images from 1 Million Identities

Arc2Face contains approximately 21 million facial images representing 1 million unique identities at a resolution of 448x448 pixels. The dataset was created by upsampling half of the WebFace42M database using a blind face restoration network for the Arc2Face foundation model research.

Use Cases

• Train ID-consistent face generation models using the 1 million distinct identity labels.
• Develop face restoration benchmarks by comparing the 448x448 upsampled images to original lower-resolution sources.
• Build large-scale face recognition systems leveraging the 21 million image samples across diverse identities.

Strengths

• Large scale of 21 million images provides ample data for model training.
• High resolution of 448x448 pixels per image, upsampled from 112x112 originals.
• Diverse coverage across 1 million unique human identities.

Limitations

• Dataset is derived from an existing source (WebFace42M), potentially inheriting its biases and limitations.
• 50% of images are upsampled, which may introduce artifacts not present in original high-quality photos.
• No information provided on demographic distribution, geographic origin, or data collection timeframe.

Provenance

Source

WebFace42M database, upsampled by the Arc2Face research team.

Collection Method

Created by upsampling 50% of WebFace42M images using a state-of-the-art blind face restoration network.

Freshness

Last updated on September 29, 2024.