TI-EDGE-AI-CLOUD: Edge AI Hello World Demo Not Working

Karthik Raveendranath

Part Number: TI-EDGE-AI-CLOUD

Hi,

I am trying to get familiarized with EdgeAI cloud tools for one of our new applications. I had followed the "Process This: Program an Edge AI "Hello World" Application Using Free Online Tools" webinar. As mentioned in the webinar, I was able to do the first 2 steps (Inference in local PC and Inference in TDA ARM controller through EdgeAI Cloud successfully). However when running the 3rd notebook (3_EdgeAI_Model_Compile_Run), no errors are thrown but inference is not happening properly. There is no overlay box in the images and the inference time/fps are entirely different from the webinar as follows.

Please help with the same.

Thanks and Regards

Karthik R

over 4 years ago

0 Srik Gurrapu over 4 years ago

TI__Expert 3353 points

Hello Karthik,

Thank you for testing this. What this means is that we need more images or more iterations for compiling so it'd be more accurate inferring the objects.

Let me check this and get back to you.

Thank you,

Srik Gurrapu

+1 Srik Gurrapu over 4 years ago

TI__Expert 3353 points

Hello Karthik,

I looked into this more. The Mobnetv1 model needs several calibration images and iterations to get better accuracy.

For faster verification of the entire flow, I updated the example to use SSD Mobilenet v2.

You can get this code from the same place from the below post.

https://e2e.ti.com/support/processors-group/processors/f/processors-forum/1014084/tda4vm-jacinto-ti-edge-ai-monthly-webinar-jul-2021-embedded-deep-learning-deployment-demystified

Please let us know if you are able to get the inference results with this.

And, for further information:

Popular models are compiled with extensive optimizations and these are available to use from the Modelzoon below. The Helloworld example is only to show a quick flow of the entire flow.

github.com/.../edgeai-modelzoo

Thanks
Srik