What is surya-ocr version

#3
by Redgalaxy2 - opened

I tested this space and the latest version of Surya-OCR (April 2025). This space performs better than the latest version. So, which version of Surya-OCR are you using for this?

surya-ocr==0.5.0

models:
https://huggingface.co/vikp/surya_layout3
https://huggingface.co/vikp/surya_order
https://huggingface.co/vikp/surya_rec2

--
I noticed the same thing. I had to clone the exact docker container locally to check on these

@Redgalaxy2 @cnmoro i tried running the latest version of surya-ocr but it is stuck at downloading the text_recognition model can you help me on this how can i test this space version locally?

@Redgalaxy2 @cnmoro i tried running the latest version of surya-ocr but it is stuck at downloading the text_recognition model can you help me on this how can i test this space version locally?

I've uploaded some prebuilt versions on docker hub, they should have the models pre-downloaded inside them.

You can try them out:

cnmoro/surya-ocr-stable (this is equivalent to this space)
cnmoro/surya-ocr-stable-minimalvram (this has some mods regarding batch sizes and so on, so it uses <4gb VRAM)

Thanks alot @cnmoro for sharing this. But i tried both of the models on my local system they are working perfectly fine. but i have captchas like this:

54.png

for which surya-ocr is also failing to provide correct answer can you suggest any alternative model or approch from which i can get the correct answer as expected for captchas like these? i tried using llm but the accuracy is really bad even for the top models from openai / gemini

@SampleUser123 you will probably have more luck using models that were trained exclusively for reading captchas, such as this: https://huggingface.co/anuashok/ocr-captcha-v3
There are a lot of options in huggingface

Sign up or log in to comment