Skip to content

algoprog/ocr

v1.1.0Apache-2.0

Reads the text in pictures, photos of documents, screenshots and scanned PDFs on this computer (RapidOCR), so agents can use what a scan or a receipt says. It reads on the graphics card when there is one (an NVIDIA GPU, or a Mac's), else on the CPU.

GitHub

Text from pictures

Reads text in pictures and scanned PDFs on this computer with RapidOCR: photos of letters and forms, screenshots, receipts, signs, and PDFs made by a scanner (pages that are pictures, with no text to select).

  • ocr_image: the text of one picture, in reading order (and each line's confidence if asked).
  • ocr_pdf: the text of a scanned PDF, page by page.

Nothing is sent anywhere. Setting it up installs RapidOCR, ONNX Runtime and pypdfium2 (about 150 MB). It reads on the graphics card when there is one: on a computer with an NVIDIA GPU, its setup also installs ONNX Runtime's GPU build with NVIDIA's CUDA and cuDNN libraries (about 2 GB); a Mac's runs it through Core ML. Otherwise, and when the card fails (it then says why, once), it reads on the CPU.