Failed to build CUDA engine for TensorRT about fast-pathology

[naguileraleal]

Today I tried running fastpathology from the terminal and making inference with the previously mentioned model using TensorRT, and it works correctly now.
I did not make any changes to the libraries I'm using.

I'm sorry but I can't explain this behaviour. I'll close this issue now. Thanks for your help!

[smistad]

Hi

TensorRT works fine on my ubuntu 20 machine, however I have older libraries installed: tensorrt 8.4, cudnn 8.4, cuda 11.2.
Could maybe be a version issue, I can test upgrading and see what happens.
Are there any other error or warning messages besides "Failed to build CUDA engine for TensorRT"? (Look in the console window where you launched FastPathology by pressing alt-tab)

Note that you can manually specify which inference engine to use in your pipeline if you want to override the selection of TensorRT. You do this by setting Attribute inference-engine OpenVINO on the SegmentationNetwork process object in the pipeline file.

[naguileraleal]

Thanks for your quick response!
I set Attribute inference-engine OpenVINO and inference started working again.

I don't see a terminal window being launched when I launch FastPathology when I click the app's icon. I did not find a command to launch FastPathology from the terminal; could you tell me how to do that?

[smistad]

Open your terminal and go to /opt/fastpathology/bin/ and run the fastpathology executable in there:

cd /opt/fastpathology/bin/
./fastpathology

Then you should get any warnings and errors in this terminal.

[andreped]

Might be that you needed to restart your machine after installing CUDA and what not. This should atleast be needed after installing drivers. Anyways. Glad its working for you!

I assume it also works fine from FP (without running it from the terminal)?