Face detection using Centerface and FaceDetectIR, built using Isaac 2020.2 and Deepstream 5.0
Start a container using the firekind/isaac:2020.2-deepstream-5.0.1-triton image (for centerface) or firekind/isaac:2020.2-deepstream-5.0.1-base (for facedetectir):
$ docker run \
--gpus=all \
--net=host \
--mount source=isaac-sdk-build-cache,target=/root \
-v `pwd`:/workspaces \
-w /workspaces \
-it \
<image tag>
Take a note of the device id of the v4l2 camera. Update the device_id in the config section of the graph (in app/graphs) which you will use.
Download the centerface model using
$ cd helpers && ./download_centerface.sh
This will download the centerface model, and updates the dimension of the input and output nodes. (In Triton Inference Server, if you want the input and output nodes to have variable size then relevant dimensions should be specified as -1. helpers/change_dim.py reads the input ONNX model, updates the height and width dimensions to -1, and saves the resulting model.)
Then, run the application using
$ bazel run //app:centerface
Download the model using
$ cd helpers && ./download_facedetectir.sh
This will download the facedetectir model from NGC, and extract the contents to model/facedetectir. Run the application using
$ bazel run //app:facedetectir
This model can be deployed on jetson as well, using the deploy.sh script.
$ ./deploy.sh --remote_user <username_on_jetson> -p //app:facedetectir_pkg -d jetpack44 -h <jetson_ip>
And on jetson, execute:
$ cd ~/deploy/face_detection_facedetectir_pkg
$ ./app/run_facedetectir
Centerface and FaceDetectIR can be run using a realsense camera as well. For centerface,
$ bazel run //app:centerface_realsense
and for facedetectir,
$ bazel run //app:facedetectir_realsense
Most of the deepstream related code for centerface was taken from NVIDIA-AI-IOT.