Jetson Nano Object Detection C/C++ Example

Robotics & Edge Computing Jetson Systems Jetson Nano
21

Hi xplanescientist, the 39FPS was using the 37-class SSD-Mobilenet-v2 model trained on the PETS dataset (see this post).

The SSD-Mobilenet and SSD-Inception models from Hello AI World are using the 90-class COCO dataset, so performance is lower with more classes. You should be able to get around ~22FPS for the network processing with SSD-Mobilenet-v2 (COCO) on Nano. Eventually, if you were to retrain your own SSD-Mobilenet model with TensorFlow and fewer classes, you could improve the performance by only training for the classes you needed.

22

Hello,

Did you find any c/c++ samples?

Thank you.

23

Hello,

I’m making a cctv camera and have a library to stream video. Can we use the jetson_inference library by calling it from our source?

Thank you.