Labeled Faces in the Wild (LFW)
Dataset composed of face photos collected on the web, tagged by name of the person. Used for facial recognition under non-binding conditions.
13,000+ JPEG images, original resolution 250x250 px (often resized to 62x47 px), RGB
GNU Lesser General Public License 3.0 (LGPL-3.0)
Description
The dataset Labeled Faces in the Wild (LFW) includes over 13,000 images of the faces of public figures, with annotations indicating the name of each individual. The images were collected from the Internet and captured under a variety of conditions, making the facial recognition task more realistic and challenging.
What is this dataset for?
- Training and evaluating facial recognition models
- Studying biometric identification in uncontrolled conditions
- Testing facial detection and classification algorithms
Can it be enriched or improved?
The dataset can be completed with more accurate annotations, additional metadata (age, expression, pose), or higher resolution images. The addition of multimodal data (video, audio) could extend uses.
🔎 In summary
🧠 Recommended for
- Biometric researchers
- ML developers
- IT security projects
🔧 Compatible tools
- OpenCV
- Dlib
- TensorFlow
- PyTorch
- FaceNet
💡 Tip
Pre-processing images to standardize lighting and pose improves model accuracy.
Frequently Asked Questions
What is the main use of this dataset?
Unconstrained facial recognition, to identify people on various images.
Does this dataset contain sensitive or private data?
No, the images concern public figures and are available on the Internet.
Can I use this dataset for a commercial project?
Yes, subject to compliance with the terms of the LGPL-3.0 license, in particular on redistribution.
