FEMNIST: Handwritten Digits and Letters for Federated Learning

FEMNIST is a dataset for image classification of handwritten digits, lowercase, and uppercase letters, providing 62 unique labels. Each sample is a 28x28 grayscale image and includes writer and character information. The dataset is part of the LEAF benchmark and was curated by flwrlabs, with a last recorded update in April 2024.

Use Cases

• Train image classification models based on the 62-class handwritten character labels.
• Benchmark federated learning systems based on the dataset's writer-specific partitioning.
• Develop writer identification models based on the included writer_id metadata.

Strengths

• Provides 62 distinct character classes for classification tasks.
• Includes writer identification metadata, enabling user-specific analysis.
• Part of the established LEAF benchmark for federated learning.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

LEAF benchmark, curated by flwrlabs.

Collection Method

Preprocessed from an original source, resembling standard image classification preprocessing.

Freshness

Last updated 2024-04-24 10:03:35; freshness should be verified.