Fidel: A Large-Scale Sentence Level Amharic OCR Dataset

Fidel is a large-scale dataset for Amharic Optical Character Recognition at the sentence level. It contains a diverse collection of Amharic text images from handwritten, typed, and synthetic sources, created by author upanzi and last updated on March 30, 2026. The dataset aims to advance language technology for Amharic in Ethiopia.

Use Cases

• Train OCR models for Amharic text based on the dataset's collection of handwritten sentence images.
• Benchmark OCR performance on typed Amharic documents based on the dataset's typed text sources.
• Develop automated form processing systems for Ethiopian applications based on the dataset's synthetic and real text images.
• Support digital ID initiatives in Ethiopia based on the dataset's focus on document digitization.

Strengths

• Dataset is described as large-scale and contains a diverse collection of text images.
• Data sources include handwritten, typed, and synthetic Amharic text, suggesting variety.
• Last updated on March 30, 2026, indicating recent maintenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and exact size are unknown, which may limit suitability assessment.
• Data may reflect geographic or source bias inherent to its collection for Ethiopian applications.

Provenance

Source

Author upanzi on Hugging Face.

Collection Method

Likely gathered from a combination of handwritten, typed, and synthetic Amharic text sources.

Time Range

null

Freshness

Last updated 2026-03-30 19:23:08; freshness should be verified.

Geography

Primarily Ethiopia, based on the described applications.