Deploy with TensorRT

[oym1994]

Hi, thanks for your great contribution. Very great job! I want to deploy it in a Jetson AGX platform. The main problem is that the inferring time is far beyond our expectation(c++ 400ms, input 500 -- 500 superpoint, the ideal time should be less than 100ms). I wander if anyone has deployed it in TensorRT platform and is it feasible to do that(all the operations are supported in current TensorRT framework)? Thanks in advance for any kind response.

[jcyhcs]

how can you convert pth model to onnx model? can you expain the detail method? thanks ! @oym1994

[Mehdi-96]

Did someone manage to convert the pth model to onnx model? I would aprreciate it if someone shares his experience

[amk2777]

Desperately need to convert this model to Tensor-RT but have not be been successful. Did anyone manage to convert the model to tensor-RT? If successful, kindly share your experience.

[Broad-sky]

Deploy with TensorRT: https://github.com/Broad-sky/feature-detection-matching-algorithm

[deblauwetom]

https://github.com/Broad-sky/feature-detection-matching-algorithm

Very nice! It would be super great if the conversion script was added 🤞 . PS On what device did you run the example? What would the performance be on the small jetson nano?

[oym1994]

I have successfully deployed with TensorRT too, with onnx model provided. Please refer to https://github.com/oym1994/ImageFeaturesAndMatching.

[oym1994]

I have successfully deployed with TensorRT too, with onnx model provided. Please refer to https://github.com/oym1994/ImageFeaturesAndMatching.

Sorry for later response! I have successfully deployed with TensorRT too, with onnx model provided. Please refer to https://github.com/oym1994/ImageFeaturesAndMatching. I hope it will still help you!