onnx Export Problem.

[GlimSH]

Hello, I studied using the project and checked the results. I tried to convert it to onnx, but there is a problem.

Is there any way to solve it?

[RizwanMunawar]

@GlimSH! Onnx conversion is not supported yet! I will add this module soon.

[JasonRuan5]

@RizwanMunawar Thanks for the great code! Just wonder if onnx conversion is ok now? We can convert pt to onnx and trt engine, but the trt engine cannot be load in C++. It would be great if it can!

[RizwanMunawar]

@JasonRuan5! At moment onnx conversion is not optimized for C++ integration. I am working on this module, but It will take time to develop. Meanwhile, you can check my Patreon subscription from the link mentioned below, which will provide you access to all my coming projects and LinkedIn project codes.

✅✅✅✅✅✅✅✅✅

https://patreon.com/MuhammadRizwanMunawarWork?utm_medium=clipboard_copy&utm_source=copyLink&utm_campaign=creatorshare_creator&utm_content=join_link

✅✅✅✅✅✅✅✅✅

Thanks.

[JasonRuan5]

@RizwanMunawar Thanks for such a quick response. Just wonder what is main difference between "converting .pt to .onnx and onnx to trt with trtexec" and "converting pt to engine directly"? Engine cannot be load if converted via onnx, but ok if converted directly. I am new to onnx and tensorrt, we are looking for sample code to segment image after loading the engine. Any suggestions/advices will be greatly appreciated!

[RizwanMunawar]

@JasonRuan5! There is no direct method for conversion of Pytorch(.pt) to Tensorrt (.engine). You will need to convert the model to onnx and then you can convert it into the trt. For jetson nano the way for conversion to (.engine) is different.

It seems that the conversion of (onnx->engine) has some issues in input shape and output shape or quantization modules etc. I am not sure at the moment what will be the best possible way because I am working on this module yet.

[madinwei]

@GlimSH! Onnx conversion is not supported yet! I will add this module soon.

Hello, @RizwanMunawar , thank you for this awesome repository and code.
is the onnx conversion ready now? I have to use c++ for my application. it will help me a lot if I can use my trained model on my data with it.
Looking forward to you reply, and Happy Ramadan

[RizwanMunawar]

@madinwei I am working on this stuff, but not been completed yet. It can take time, but I will try to add this module as soon as possible. Thanks.

[RomeroRomeroMartin]

Hello @RizwanMunawar, where can I find information to do the onnx conversion for my own, until your module is not available?

Thanks

[Sameersingh79]

@RizwanMunawar Any update on the onnx conversion??