Detectnet_v2 - Message type "Inference" has no field named "evaluation_config"

[bke]

Hi, a question - I’m trying to run inference using tao-deploy for the detectnet_v2 model using an engine file. This gives the following error:

…
Traceback (most recent call last):
File “</usr/local/lib/python3.8/dist-packages/nvidia_tao_deploy/cv/detectnet_v2/scripts/inference.py>”, line 3, in
…
google.protobuf.text_format.ParseError: 1:1 : Message type “Inference” has no field named “evaluation_config”.

What would be really helpful is if the “Inference” message definition is printed. Now I have to guess if perhaps the documentation that I am following is out-of-date, or there might be a bug, or I am still doing something wrong after trying for one hour. I tried finding the source code to see what is going on - but the pip package nvidia-tao-deploy is obfuscated! Why? This makes it really hard to tackle the issue without your help :-/

I can provide you with the full details like the spec file and everything, but I would rather be able to solve this issue myself.

Q1 - Can you provide the nvidia-tao-deploy without obfuscation?
If not, can you provide the “Inference” message proto?
If not, what do you suggest?

[Morganh]

@bke

Hi, a question - I’m trying to run inference using tao-deploy for the detectnet_v2 model using an engine file. This gives the following error:

…
Traceback (most recent call last):
File “</usr/local/lib/python3.8/dist-packages/nvidia_tao_deploy/cv/detectnet_v2/scripts/inference.py>”, line 3, in
…
google.protobuf.text_format.ParseError: 1:1 : Message type “Inference” has no field named “evaluation_config”.

I am afraid there is something mismatching in the inference spec file.
Please refer to the notebook and find the detectnet_v2’s spec file and its .ipynb file.

You can download via

wget --content-disposition https://api.ngc.nvidia.com/v2/resources/nvidia/tao/tao-getting-started/versions/4.0.12/zip -O getting_started_v4.0.2.zip
unzip -u getting_started_v4.0.2.zip  -d ./getting_started_v4.0.2 && rm -rf getting_started_v4.0.2.zip && cd ./getting_started_v4.0.2

[bke]

Thank you for the fast reply.

I am basing my config on:
https://docs.nvidia.com/tao/tao-toolkit/text/tao_deploy/detectnet_v2.html#running-inference-through-tensorrt-engine

The example you show does not use tao-deploy with a .engine file but the tao command with the .tlt file.

However:

1. Using the documentation I was able to create a new spec file and run inference.
2. Given the new spec file this also works for the tao-deploy command

Conclusion:
The documentation that I was following is incorrect. The correct syntax is in the provided examples.

[Morganh]

Glad to know it is working now. The link you mentioned above is for evaluation instead of running inference.

Yes, please refer to the inference spec file in the notebook.