Sudhir Kshirsagar
Created December 4, 2019 © GPL3+

Retinal AI

Detection of diabetic retinopathy and retinal microaneurysm using NVIDIA Jetson Nano with direct image capture and edge AI.

IntermediateFull instructions provided16 hours37
Retinal AI

Things used in this project

Hardware components

NVIDIA Jetson Nano Developer Kit
NVIDIA Jetson Nano Developer Kit
×1
Logitech USB Webcam
×1

Story

Read more

Schematics

Overall setup

The laptop was used to display the retinal images. Those were captured with the webcam (mounted on the box that the Nano came in!). The images were captured using the Jupyter notebook, and used the train the model. Then the trained model was used to evaluate additional images from the laptop. Eventually the laptop/webcam will replaced by a small wearable eyepatch that will capture the retinal image.

The Nano

Powered by an adapter per the conventional wisdom

The image capture setup

The laptop displayed an image, and the webcam was used to capture it.

Output in the monitor attached to the Nano

The predictions were generated using the Python script. There was a slight lag between the actual capture and the image showing up in Jupyter notebook. The overall accuracy of the model trained with the miniscule dataset was reasonably good.

Code

Jupyter Notebook that is part of the Getting Started with Nano DLI course

Python
The existing notebook that recognized a thumbs up image was modified to handle the five new categories.
The key was to start with the Nano image that is provided by Nvidia DLI. That came configured with all the necessary tools.
# the interactive classification Jupyter notebook from NVIDIA DLI was used with modifications.

Credits

Sudhir Kshirsagar

Sudhir Kshirsagar

4 projects • 2 followers

Comments