SageMaker Object Detection training

Make sure you've seen this if you need help creating a dataset first! If you already have a usable dataset, follow along here to train a model:

Create a training job

Navigate to SageMaker console and click on “Training jobs”. Once there, click on “Create training job”.

In the job details, add the job name, create a new IAM role or use an existing role which has necessary permission. Select “Object Detection” algorithm from the drop down and use “Pipe” as the Input mode.

Continuing on training job details, put appropriate instance type and maximum runtime.

In the hyperparameter selection section, make sure to put correct number of training samples.

Edit input data configuration section and make sure to put S3 location for the output.manifest file that was created from Ground Truth labeling job.

Create another channel named “validation” and fill in similar details as test channel.

At last, specify the S3 path to save output model artifacts. Click on “Create training job”

You will see the training job in progress on your SageMaker console.

Once the training job is finished successfully, you will see the status as “Completed”. We will use the output artifacts to create a deployable model, click on “Create model”.

In the create model section, add model name and verify the auto-populated details. Click on “Create model”.

Once the model is created, you can click on “Models” from SageMaker console and see its summary