Replicating the huggingface repo #218
Replies: 1 comment 2 replies
-
Thank you for your kind words. Regarding your question concerning HF space (duplicating, changing hardware, etc), I am unable to provide an answer and I could only point to the docs available for HF Spaces in general. You are right, the deepdoctection space has better private models and AWS Textract as powerful OCR engine but DD_ADDONS itself does not really add a lot of magic. It only has a commercial PDF mining tool that is only needed to reduce Textract costs and a few NMS steps thereafter. I do not even think that the NMS post processing does still add any value, leaving it for historical reasons. |
Beta Was this translation helpful? Give feedback.
-
Hi @JaMe76,
First I would like to thank you for building this amazing repo.
Question: I would like to understand how I can replicate the huggingface space for private use (for a business use case).
Context: Our customer would like to replace manual checking of an FMCG product packaging (eg: Cheetos cover) for text information. So running it locally or as a huggingface space with GPU enabled would be ideal.
Good to have: I understand that it currently runs on CPU and sub 30 seconds results are impressive. But is it possible to run the same space with GPU and achieve faster results ?
What I have tried so far: I was able to run deepdoctection on an AWS EC2 machine. But I believe some parts of what makes the huggingface space work so well (like DD_ADDONS, Fully trained model) are not available, so the results are a bit underwhelming.
I am new to github discussions so please let me know if this is a bit out of scope.
Beta Was this translation helpful? Give feedback.
All reactions