What is the Clara AGX Dermatology Reference Application?

The Dermatology Reference Application is an example of how one can build a simple application around computer vision models on the Clara AGX platform.

The application consists of two computer vision models - the first is a YoloV4 object detection model that looks for moles from a webcam feed, and the second is an EfficientNet B0 classification model trained on the HAM10000 dataset, that classifies a mole as either: Melanoma, Benign, Nevus, or Unknown.

The models were trained by Roujia Wang (roujia.wang@duke.edu)

Requirements

Either a Jetson Xavier AGX device, or the Clara AGX Developer Kit running Jetpack 4.5 or later.

How to run the application

To run:


sudo docker pull nvcr.io/nvidia/clara-agx/agx-dermatology-reference:v1.1


export DISPLAY=:1


xhost +


mkdir /home/nvidia/results


sudo docker run -it --rm --net=host --runtime nvidia -e DISPLAY=$DISPLAY --device /dev/videox:/dev/video0 -v /home/nvidia/results:/results nvcr.io/nvidia/clara-agx/agx-dermatology-reference:v1.1


cd source


python3 demo.py


License

Licenses and model files are available. They can be pulled as part of the procedure described. By pulling and using the container, you accept the terms and conditions of these licenses.

Suggested Reading

For documentation, SDK, and how to get a Clara Developer Kit: https://developer.nvidia.com/clara-holoscan-sdk

Use the NVIDIA Devtalk forum for questions regarding this Release: https://forums.developer.nvidia.com/c/healthcare/clara-holoscan-sdk/