Synthetic Receipt Images and Text for OCR Training

Synth Receipts Ocr provides a dataset of synthetic receipt images paired with text for optical character recognition tasks. The dataset, created by sushanthr, was last updated in March 2026. Platform tags indicate it contains at least 100,000 records and is optimized for processing with libraries like Polars and Dask.

Use Cases

• Train an OCR model to extract text from synthetic receipt images.
• Benchmark text detection and recognition algorithms on structured document layouts.
• Fine-tune a multimodal model for joint image-text understanding of receipt data.
• Develop data augmentation pipelines using synthetic receipt variations.

Strengths

• Contains at least 100,000 records as indicated by the '100 Kn1 M' size category tag.
• Data is stored in an optimized Parquet format for efficient loading with libraries like Polars and Dask.
• Dataset is multimodal, combining image and text modalities for comprehensive OCR tasks.

Limitations

• The synthetic nature of the data may not perfectly capture the noise and variability of real-world receipts.
• Specific column names and data structure are unknown, complicating initial exploration.
• Geographic coverage is suggested as 'us' but the exact diversity of receipt formats is unclear.

Provenance

Source

Hugging Face user sushanthr.

Collection Method

Synthetically generated, method unknown.

Freshness

Last updated on 2026-03-09.

Geography

Region tag suggests a focus on 'us' (United States).