Selection of Jetson processor for ML/AI application

Hi,

We are working on defect detection project. At present we have identified few models for defect detection, these models are tested in Google colab and are working as expected.

When we tested the models in a local Linux PC, few models are taking more time for prediction when compared to Google colab. For the final installation we are planning to deploy the model in a SBC running Linux.

Based on the testing results, we understood that SBC should have GPU or dedicated SBC to be selected for final installation, while searching for suitable SBC we found Jetson will work for our application but, we are not able to identify the suitable product (Jetson Nano/Jetson TX2/Jetson Xavier NX) for our application.

We would like to which product will suit for ML/AI application.

While going through the authorized distributor website it’s not clear whether we need to buy the carrier board and processor board separately or the processor boards is included when we buy the carrier board. Can you please share the link of authorized distributor (Authorized distributor from India) from whom we can buy carrier board and processor board as a single package.

Thanks,
Sreekanth

Hi,

You can check below benchmark table to find one for your use case:

Thanks.

The models which we have trained and tested in Google colab are,

  1. Yolov5
  2. Alexnet
  3. Squeeznet

So I presume these models can work on the Jetson processor. Our main concern is the time taken for prediction. When running the models in Google Colab, the time taken for prediction (Image as input) is approximately 200ms.

The processor (Nano/Xavier/TX2) should not take more time when an image is passed to the model for prediction.

Based on your experience can you let us know which processor we can go with.

Thanks,
Sreekanth