Integrated Jetson nano + dongle coral edge tpu

[Python Developer]

Hello guys.
I would need some advice regarding a new project I have to develop for a company. but I need some acceptable advice before start the project.

Q1- I need to know is it possible to use one jetson nano with some dongle coral edge tpu? is it efficiet way for product?

My project is running multiple deep models on device and give me 2-5 FPS for multi-stream.
The models included face recogntion and object detection , …
I get 25-30 FPS for object detection on jetson nano, perhapse I get ~10 fps for this task in multi-stream. but I want to run some other models along with this model.

one possible way is that to use multi jetson nano. but this way is Affordable in product, because each jetson nano cost is 100$.

Q2- other solution can be integration multiple dongle coral edge tpu to jetson nano. I don’t know, Is this way is hard? are there documets for this way?

Q3- other sultion can be integration multiple dongle coral edge tpu to coral edge tpu dev board? but dev board is a bit expensive than jetson nano.

[Sayak Paul]

Hi,

I am Sayak, working at PyImageSearch as a Deep Learning Associate. At PyImageSearch, we are pretty much invested in using both Jetson Nano and Coral TPU USB Accelerator. Here are some pointers:

• In our experience, a combination of RPi 4B (4GB) and the USB Accelerator stood out to be the best one given our requirements. It's possible although might be a bit more involved to plug in the USB Accelerator with a Jetson Nano. Know more here: https://forums.developer.nvidia.com/t/m-2-pcie-google-coral-in-ubuntu/108963/4. 
• I don't think this would be a good idea because then you might be restricting the full potential of the Nano board since it is capable of running optimized inference with TensorRT. 

Sayak Paul | sayak.dev

--
You received this message because you are subscribed to the Google Groups "Discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to discuss+u...@tensorflow.org.
To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/discuss/f36e96ba-f788-4b78-a7cb-243c9b12661b%40tensorflow.org.

[Python Developer]

In the RPi4CV of pyimagesearch book, one chapter is related to running image classification and object detection on Coral, in this chapter,  Adrain use RPi along with Coral? Is it possible to use these codes for jetson nano anlog with Coral?

I don't understand your last advise.

To unsubscribe from this group and stop receiving emails from it, send an email to dis...@tensorflow.org.

[Sayak Paul]

We do use it. Those codebases won't run on Jetson Nano though. We have separate chapters on Jetson Nano in the book. 

I don't think the idea of using the USB Accelerator with the Nano would be a good idea because then you might be restricting the full potential of the Nano board since it is capable of running optimized inference with TensorRT. 

Sayak Paul | sayak.dev

To unsubscribe from this group and stop receiving emails from it, send an email to discuss+u...@tensorflow.org.
To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/discuss/5c023402-7bd0-46c3-8192-b40348d48810%40tensorflow.org.