Face Recognition Deepstream App

NVIDIA Jetson Xavier NX
DeepStream Version5.0
Jetpack 4.4.1 [L4T 32.4.4]
CUDA: 10.2.89
OpenCV: 4.1.1 compiled CUDA: NO
VPI: 0.4.4
Vulkan: 1.2.70

I am a newbie to Deepstream and face recognition technology overall. I have a project i want to work on, it is about Face Recognition System using Deepstream.
It includes 3 parts:
*Facial recognition
*Tensorrt Pruned and Quantization

Please can you guide where i can start from? I am will to learn everything related to this project. Thank you in advance.


Do you have a face recognition model?

Usually, this includes a detector and a classifier.
You will need to first locate a face within an image, and then recognize the identity.

We don’t have any face related model in our deepstream SDK.
So please start with training a model with the TLT toolkit and your own face database:

TLT can also help you do the pruning.
After that, please following below sample to deploy it with Deepstream.


Thank you a lot.

About the facial recognition model, which one do you recommend? I want to use an .mp4 video file as a test file.


The choice is a tradeoff between performance and accuracy.
Usually, you can check the table and find the most accurate one under the acceptable fps.


i dont mind about performance when it is more than enough accuracy

is facenet_keras.h5 supported by TLT?

1 Like


Sorry that only NVIDIA pre-trained models from NGC are currently supported.