Labeled Faces in the Wild
A classic benchmark dataset of thousands of labeled face images collected from the web, designed for unconstrained face recognition research and commonly featured in awesome machine learning dataset collections.
About this tool
Labeled Faces in the Wild
Overview
Labeled Faces in the Wild (LFW) is a widely used benchmark dataset for unconstrained face recognition research. It consists of thousands of labeled face images collected from the web and is considered a de facto standard test set for evaluating face verification and identification algorithms.
- Type: Dataset
- Category: Datasets
- Domain: Computer Vision, Machine Learning
- Primary Task: Face verification and face recognition in unconstrained settings
- Provider / Origin: University of Massachusetts Amherst
- Official Site / Documentation: http://vis-www.cs.umass.edu/lfw/
Features
-
Dataset Scale
- Contains more than 13,000 labeled facial images.
- Images correspond to 1,680 different people.
-
Data Source & Collection
- Faces are collected from the web, reflecting natural, real-world image conditions.
- Images are “in the wild” (unconstrained), rather than studio or controlled settings.
-
Task Orientation
- Designed primarily for face verification (deciding whether two images are of the same person).
- Commonly used as a benchmark for face identification experiments as well.
-
Benchmark Status
- Described as the de facto academic test set for face verification.
- Frequently used to report and compare performance of deep learning models (e.g., FaceNet).
- Well-established in research literature and “awesome machine learning dataset” collections.
-
Typical Usage in Research
- Evaluation of deep face embedding models (e.g., FaceNet) via verification accuracy.
- Threshold-based verification experiments where Euclidean distance between embeddings is used as a similarity measure.
- Comparison and benchmarking of new face recognition systems and architectures.
Example Benchmark Reference (from content)
- FaceNet reported:
- 98.87% ± 0.15 accuracy on LFW in one setting.
- 99.63% ± 0.09 accuracy on LFW in another setting.
- An example optimal verification threshold around 1.24 (on embedding distance), illustrating common LFW usage.
Tags
- datasets
- computer-vision
- machine-learning
Pricing
- Not specified in the provided content. Refer to the official LFW website for license and access details.
Loading more......
Information
Categories
Tags
Similar Products
6 result(s)A facial age and gender estimation dataset with approximately 375k images of famous figures, biometrically filtered to improve label quality. Indexed within an awesome machine learning datasets collection.
A set of context-aware recommendation datasets across five domains, distributed with CARSKit, for research in context-aware recommender systems and machine learning. Part of an awesome public datasets listing.
Public click-through and display advertising dataset released by Criteo for CTR prediction research, widely used in machine learning benchmarks and included in awesome advertising/clickstream datasets lists.
A collection of standardized datasets for classification and regression tasks maintained by the University of Toronto’s DELVE project, widely used for benchmarking machine learning algorithms and referenced in awesome dataset directories.
A large-scale dataset of Farsi and Arabic license plate characters for character recognition and license plate recognition research, registered within an awesome public machine learning datasets collection.
A curated awesome-style collection of image processing and computer vision datasets, hosted under the Awesome Data (apd-core) project. The listed datasets (e.g., ImageNet, KITTI, Danbooru, DukeMTMC) are part of this meta awesome directory of specialized data resources.