J742S2XH01EVM: Cpp Inference Examples

Danilo A.

Part Number: J742S2XH01EVM
Other Parts Discussed in Thread: AM69A

Hi Team,

Posting on behalf of our customer.

I've been compiling the CPP examples provided by edge-ai-tools. However, I can't get them to work on the board. When I run `/tidl_classification`, for example, I get an OpenCV error in the `resize` function.

On the other hand, I've tried to create my own inference examples. I was able to do it in Python because the `onnxruntime-tidl` SDK includes the `TIDLExecutionProvider`. However, CPP doesn't include the provider, which is why the examples use the `itidl_rt.h` library directly, instead of working with `onnxruntime` as in Python. I need to know if it's possible to add the `TiDL Execution` provider to the CPP `onnxruntime`, for example.

Regards,

Danilo

3 months ago

0 Wen Li 3 months ago

TI__Expert 8081 points

Hi;

I would suggest to get the examples running on the board first. Then we can look into your inference case.

I have tried the examples in the link below, they worked fine.

https://github.com/TexasInstruments/edgeai-tidl-tools

First of all, you will have to make a boot-able miniSD card and compile the examples to generate artifacts(model-artifacts),

# Clone a "edgeAI TIDL tools" on a linux PC and copy the entire directory to EVM
scp -r edgeai-tidl-tools/ root@128.247.79.115:/opt/.

#copy both "model-artifacts" and "models" directories to EVM
scp -r model-artifacts/ root@128.247.79.115:/opt/edgeai-tidl-tools/.
scp -r models/ root@128.247.79.115:/opt/edgeai-tidl-tools/.

#go to EVM and cd /opt/edgeai-tidl-tools/
git checkout 11_xx_xx_xx
export SOC=am69a
rm -rf build bin lib
mkdir build
cd build
cmake ../examples/
make
cd ..
python3 ./scripts/gen_test_report.py

After a successful run, you should be able to see the inference results in the output_images and output_binaries.

Best regards

Wen Li

0 Danilo A. 3 months ago in reply to Wen Li

TI__Mastermind 31760 points

Hi Wen,

Thank you, thats great this response, but i have other question. I would know how to deploy the C7x consumption. I don't know the registers or IP to read the consumption of C7x cores, similar to this image but we don't need create an interface, only follow the consumption. Thank you.