This repository contains example notebooks to train and compile deep learning models using Amazon Sagemaker for ai-vision-solution-kit.
The examples demonstrate training followed by deep learning model optimization for an edge device. Similarly, the examples provide you a gentle introduction to Sagemaker functionalities. Amazon Sagemaker facilitates training and compiling models with different algorithms and frameworks. Using Amazon Sagemaker Python SDK, we can write code to train MXNet, Tensorflow, Pytorch framework models. Similarly, the examples explain the EC2 instances for the training task and the cloud storage service S3 bucket which provides better performance thus making the training task easier for a user.
Sagemaker Neo compiles the trained machine learning model to obtain optimal performance on the target device.
The notebook samples contain fine-tuning pre-trained object detection models. Then the trained models will be compiled for deployment on Jetson Nano.
-
Mask Detection SSD Mobilenet GluonCV - The example describes fine-tuning an object detection model that detects face masks. Single Shot multibox Detection(SSD) with MobilenetV1.0 is used for object detection.
-
Office Items Detection SSD Mobilenet GluonCV - Finetune a GluonCV SSD MobilenetV1.0 model to detect the office items such as Computer keyboard, Computer mouse, Computer monitor, Laptop, Desk and Chair.
Examples that offer a gentle introduction to compiling existing models for Jetson Nano.
- Person Detection SSD Mobilenet GluonCV - This example focuses on compiling an existing model in GluonCV. The model detects persons in an image.
For beginners, please refer to the setup guide for ai-solution-kit, training, and compiling model for the target device and finally deploying the model on the edge device AI Vision Solution Kit.