xView3 SAR - Object detection on giga-pixel satellite imagery

Overview

Build a scalable machine learning pipeline using Amazon SageMaker to train and deploy an object detection model for gigapixel-scale satellite imagery

In this repository, we use Amazon SageMaker to build, train, and deploy a custom Faster-RCNN model for gigapixel-scale satellite imagery. The custom FRCNN model is built using Detectron2, an open-source object detection library released by Meta AI Research.

This repository shows how to do the following:

• Build Docker images and push them to Amazon ECR to run Amazon SageMaker jobs for processing, training, and runnng inference on the xView3 SAR dataset.
• Run Amazon SageMaker Training job to preprocess xView3 SAR satellite imagery for ob
• Customize the FRCNN model forRun Amazon SageMaker Training job to finetune pre-trained model weights on xView3 SAR satellite imagery.

Installation Instructions

• create virtual environment:

python3 -m venv venv-xview3
source venv-xview3/bin/activate
python3 -m pip install --upgrade pip

• install dependencies:

pip install -r requirements_cpu.txt

OR

pip install -r requirements_gpu.txt

• install detectron2:

pip install git+https://github.com/facebookresearch/detectron2.git

• install repository as package:

pip install -e

Security

See CONTRIBUTING for more information.

License

This library is licensed under the Apache 2.0 License. See the LICENSE file.