Using the built-in image classifier
Amazon SageMaker’s built-in algorithms and pre-trained models address everyday use cases such as image classification, text summarization, and anomaly detection. In this section, you’ll recreate the famous Hot Dog, Not-Hot Dog
classifier using MXNet by following these steps:
- Upload the dataset to Amazon S3.
- Create the training job definition.
- Run the job.
- Verify the results.
Upload the dataset to Amazon S3
This chapter’s repository includes the hotdog-nothotdog
image set. Upload these files into an Amazon S3 bucket:
$ aws s3 sync 10_SageMakerModel/hotdog-nothotdog s3://ch10-cv-book-use2/sagemaker/hotdog-nothotdog --region us-east-2
Prepare the job channels
Amazon SageMaker uses channels
to sequence the training, testing, and validation data into the models and algorithms. The channel’s format will vary from simple delineated manifest files (e.g., using tabs or commas) to RecordIO...