Multiple models with Jetson Inference

Autonomous Machines Jetson & Embedded Systems Jetson Nano
1

I want to recognize the people, the faces and the joints of the body. All this is possible, but with different models, if I’m not mistaken.

My question is, if I have this code that detects objects well:

As I can make it detect the face and the joints of the body.

Context:
I am new to jetson nano, I want to create a code to keep it always running on my Jetson Nano, with Python. That is, sometimes I will use only one model, sometimes I will use another, sometimes both together. Going in and out of the code to change the model is very slow. I’m going to create a database to control the While conditionals, that is, if X table is false, the model is deactivated, if it is true, the model is activated.

3

Hi,

Would you mind checking our Deepstream SDK first?
It allows multiple inferences in the same pipeline and it might meet your requirement:

Thanks.

1 Like
4

Hello, thank you very much for your answer.

I’m honestly looking for a quick and simple solution (I haven’t analyzed what you sent me yet). I thought about using this python plugin: face-recognition · PyPI

I have not tried it yet, but I would like to know if you think it is the best or if it is possible.

They mention it in this post: Face recognition first call very slow on Jetpack 4.6 - #4 by jc5p

5

Hi,

It has some performanc issue since the camera interface is not well-optimized for Jetson.
Thanks.

1 Like
7

I am interested in using 3 models: pose, objects, faces.
The one you recommended: deepstream-sdk has no poses.

I repeat my question: is it possible to use the 3 models of detecting objects, poses and faces with jetson inference?

Btw, Thanks once for your reply.

8

Technically you can do this (if you have enough memory, which is not certain on Nano), just load the multiple models like you see in detectnet.py / posenet.py and combine those parts into one script. You will want to do your own command-line parsing for the model loading or avoid that, since the command-line parsing in jetson-inference is setup for just one model.

Also, you may be able to use just posenet for your uses, because it can provide you bounding box of the whole human, and you can also look for the face parts of the pose. That would be a lot faster than running three separate models to accomplish essentially the same thing.

11

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.