Face recognition using dlib with Jetson nano

Hi,
I am using the following tutorial to implement face recognition on Jetson nano
post link

The tutorial uses Dlib and face_recognition library to do the task of face detection and recognition. it fetches 6-7 FPS on Jetson nano, how can I incorporate this tutorial with deepstream as i want to run it for several camera streams.

thanks

Hi,

dlib has its own detection and recognition model.
If you can convert the model into TensorRT, you can just replace the path and parser in a standard pipeline:

If the model doesn’t be fully supported, you can try to insert the dlib implementation into the below sample.

Please noted that the buffer transfer between components is a GPU buffer.
Please check if dlib can accept the GPU buffer or not first.

Below is the TensorRT support matrix for your reference:
https://docs.nvidia.com/deeplearning/tensorrt/support-matrix/index.html

Thanks.

Thanks @AastaLLL, I’ll check all the methods you mentioned and get back. Thanks once again

is there any already existing pipeline for face recognition with deepstream?
thanks

Is there a way to just get a video feed with deepstream from multiple sources, I don’t want to run any inference.

Hi,

We don’t have a face recognition pipeline since there is no face related model in the Deepstream.

To feed multiple sources into Deepstream is easy.
For example, with /opt/nvidia/deepstream/deepstream-5.0/samples/configs/deepstream-app/source30_1080p_dec_infer-resnet_tiled_display_int8.txt.

1. Turn off inference

...
[primary-gie]
enable=0
gpu-id=0
...

2. Run

$ deepstream-app -c source30_1080p_dec_infer-resnet_tiled_display_int8.txt

Thanks.

Awesome, thanks. how can i do the same in python @AastaLLL

where can i access the frames of this stream @AastaLLL

Hi,

You can find some python-based sample in our GitHub below:

To access the image data, please check the following comment for information:

Thanks.

Thanks @AastaLLL I’ll look into it.

how can i disable pgie and tracker in the following example
thanks

Hi,

You can comment out this line:

Thanks.

Thanks @AastaLLL