How to Add Open-Source OCR Models to Foundry?

[fac7a5bb66119b7b9c34]

Hello
While considering a better parsing model than what Foundry provides for document parsing, I discovered that the local model called paddleOCR(https://github.com/PaddlePaddle/PaddleOCR) is suitable for my environment. Based on the English model, it’s about 10MB. Can I load this model into Foundry for use? If so, could you please provide detailed instructions on how to do this?

[william]

@fac7a5bb66119b7b9c34

While considering a better parsing model than what Foundry provides for document parsing, I discovered that the local model called paddleOCR(https://github.com/PaddlePaddle/PaddleOCR) is suitable for my environment. Based on the English model, it’s about 10MB. Can I load this model into Foundry for use? If so, could you please provide detailed instructions on how to do this?

Hi,

You can upload custom models like the one you found online to Foundry, but we have a new product build around this use case: AIP Document Intelligence. Docs: https://www.palantir.com/docs/foundry/document-intelligence/overview

This should allow you to test PDF inputs with different OCR and LLM extraction techniques, and deploy to a transform with your custom prompt and media set. Also since this product is Beta, there might still be some minor issues, but let us know if you have any feedback!