SROIE 2019 Text Recognition: 52,330 Cropped Receipt Images for OCR

52,330 cropped text images derived from 973 scanned English receipts in the SROIE dataset. The dataset was prepared by user priyank-m and uploaded to Hugging Face in August 2022. It is split into 33,626 training images and 18,704 test images, with text annotations stored in a metadata.jsonl file.

Use Cases

• Train text recognition models based on cropped receipt images.
• Benchmark OCR accuracy on real-world scanned documents.
• Develop information extraction pipelines for receipt data.
• Fine-tune pre-trained vision-language models on domain-specific text.

Strengths

• Contains 52,330 total annotated images, providing substantial data for model training.
• Offers a clear train/test split with 33,626 and 18,704 images respectively.
• Based on the established SROIE dataset, which contains 973 source receipts.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last updated 2022-08-27 21:38:24; freshness should be verified.
• Data may reflect the specific characteristics and potential biases of the source SROIE receipts.

Provenance

Source

Derived from the Scanned receipts OCR and information extraction (SROIE) dataset.

Collection Method

Cropping bounding boxes from scanned receipts to generate individual text images.

Freshness

Last updated 2022-08-27 21:38:24.