AWS Panorama Custom Object Detector Example (Pikachu Detection)

This is an end to end example that shows how to create a custom Object Detector using GluonCV / PanoramaSDK and using it on the Panorama device

Files Included

Lambda (Folder)
- PikachuDetection.zip
Notebook(Folder)
- pikachu_detection_custom_object_detector.ipynb

Use Case

Build a custom object detector that uses transfer learning to detect Pokemon (pikachu) in an image
Details how to loop in annotations, creating the model and exporting the hybridized model
The lambda also counts the number of pikachu detected in the frame and displays it on the output HDMI

How to use the Notebook

The included Jupyter Notebook gives a helpful introduction of

Task at hand
Step by step walk thru of how to train your own Object Detector
Exporting the model and creating a tar.gz file with the parameters
Upload the model to S3 bucket
Create and publish a lambda function with the included Lambda zip file

Example Output From Notebook

An example output display is shown here

How to use the Lambda Function

The included Lambda function is a zip file that can be directly uploaded to the Lambda console to create a usable Lambda arn.

Other resources to use