[offtopic] make a reimplementation in C++

[Kromtar]

I would like to know if the community that uses this repository would be interested in a C++ reimplementation. This OCR has given me excellent results so far. But the fact that it is in Python and uses Torch makes it difficult to compile it in an optimized way. I am planning to make an exact reimplementation of the current version of EasyOCR but in C++ and using TorchScript. It is a lot of work. Is anyone else looking for a solution like this?

Greetings

[euirim]

I'm interested, particularly for inference. I don't have great C++ skills, but I may be able to write a bare bones Rust implementation that can compile into a C library.

[cfanatic]

I am interested, too.

[Cilouche]

I'm interested, thanks

[BBO-repo]

Totally agree, easyocr give great results thanks so much to the author and contributors

Having a C++ implementation would be great. I'm coding in C++ for more than 5 years @Kromtar please let me know if I can help

[yunus-temurlenk]

count on me

[kevinkazzo]

I started this, anyone else start?

[BBO-repo]

On my side start working on translating the craft algorithm (used by easyocr for text detection) in C++.
If successful I'll try to go on with translating the text recognition algorithm

@kevinkazzo if you want we can join our effort?

[facug91]

It would be great if someone could do some kind of planning so that everyone willing to help can address those different issues without overlapping each other. I know it's a lot of work I'm proposing for the person who would do it, but it would help a lot to speed up development.

[BBO-repo]

Maybe this repository is a good starting point
https://github.com/jackvial/tuatara

[kevinkazzo]

sorry I forgot to update. I did this, will respond on my other github and upload the project and start allowing people to make PRs to improve it. Did not optimize it entirely, but got the detector and recognition network working in plain torchscript/ c++. Even got it in ONNX

[BBO-repo]

Great!
Looking forward to seeing it and testing.

[MinGiSa]

sorry I forgot to update. I did this, will respond on my other github and upload the project and start allowing people to make PRs to improve it. Did not optimize it entirely, but got the detector and recognition network working in plain torchscript/ c++. Even got it in ONNX

When can we expect to see your amazing project? :)

[SC-One]

[C++ software engineer]
you can count on me too.

[ksasso1028]

Hey all, this is @kevinkazzo, on my main account will post it up by tomorrow evening!

[ksasso1028]

hey all! uploaded here it is! - > https://github.com/ksasso1028/EasyOCR-cpp

[A2va]

For those who don't want to use libtorch, I implemented easyocr with opencv, onnxruntime and xtensor. This is not a full implementation, but I will include it in a library of mine that will come out later, for now it is just this.

The ONNX model is converted with the notebook provided by samiechan here

If you have any problems or suggestions about my code, I'm happy to listen to your suggestions/questions. I will edit this message to add CRAFT model when it's ready.

https://gist.github.com/A2va/64eea11018192284c7ce3db7bb3801bc

[BBO-repo]

Thank for those sharing, I'll give it a try!