Getting Started with YOLTv4 for Object Detection in Imagery: Getting Training Data

This is the second article in a series of tutorials that attempt to fill gaps in this excellent article: Announcing YOLTv4: Improved Satellite Imagery Object Detection. The first article, covered building a container to run the model. In this article, I cover how to get the training data.

Getting Training Data

We need to train the model with a data set. The article points to the RarePlanes data set, which is an open-source dataset provided by In-Q-Tel's CosmiQ Works. The data is stored in AWS bucket and accessible with the AWS CLI. The first step to getting the data is install the AWS client.

Get AWS CLI

To install on macOS, you can use the brew package manager:

$ brew install awscli

or you can download the GUI installer.

To install on Linux:

$ curl "https://awscli.amazonaws.com/awscli-exe-linux-x86_64.zip" -o "awscliv2.zip"
$ unzip awscliv2.zip
$ sudo ./aws/install

To install on Windows, download the installer and follow the installation prompts.

Downloading the training data

The training data doesn't require an AWS account, and you can see the contents of the S3 bucket:

$ aws s3 ls s3://rareplanes-public/ --no-sign-request

                           PRE real/
                           PRE synthetic/
                           PRE weights/
2020-06-09 08:27:59      20605 LICENSE.txt

To download the data set use the sync command.

NOTE: The RarePlanes training data is over 500GB, make sure you have sufficient storage.

$ aws s3 sync s3://rareplanes-public/ --no-sign-request

Depending on the speed of your Internet connection, this can take some time.

Next step: Training the model

In the next article I'll go through the training steps in the python notebook in the YOLTv4 repo.