Rekognition: Create dataset from the CLI

Part 991 of AWS from Zero. This is lesson 9 in the Rekognition track.

What we are learning

Use create-dataset to create or register one Rekognition resource through the CLI. This lesson identifies the required input shape, saves the raw response, and keeps inspection separate from execution.

The AWS CLI operation is aws rekognition create-dataset. Required operation inputs: --dataset-type (string), --project-arn (string). The modeled top-level response contains DatasetArn.

Before you run it

aws sts get-caller-identity
REGION="${AWS_REGION:-ap-south-1}"
DATASET_TYPE="replace-with-dataset-type"
PROJECT_ARN="replace-with-project-arn"
aws rekognition create-dataset help

Use a sandbox account or an approved learning environment. Read the operation help before supplying identifiers, ARNs, network ranges, policy documents, or customer data.

Cost note: Images, video minutes, face metadata storage, and custom labels can incur charges.

The command

aws rekognition create-dataset \
  --dataset-type "$DATASET_TYPE" \
  --project-arn "$PROJECT_ARN" \
  --region "$REGION" \
  --output json > part-991-response.json

The response is saved to part-991-response.json so inspection is separate from execution. The explicit variables above keep required identifiers visible before the API call.

Inspect the result

node -e "const r=require('./part-991-response.json'); console.log(Object.keys(r))"
node -e "const r=require('./part-991-response.json'); console.log(JSON.stringify(r, null, 2))"

Compare the returned identifiers and status fields with the account, Region, and resource you intended to target. For asynchronous operations, continue with the service's matching get, list, or describe command until it reaches a terminal state.

One tiny variation

node -e "const r=require('./part-991-response.json'); console.log(JSON.stringify(r["DatasetArn"], null, 2))"

This variation changes output inspection rather than adding another infrastructure concept. Keep the raw JSON while developing a query so a narrow projection does not hide an error or unexpected field.

Common mistake

Do not run a generated request unchanged. Replace every placeholder, add ownership tags where supported, estimate cost, and verify that the selected Region and account are disposable.

Cleanup

# Review the inverse operation before removing the demo resource.
aws rekognition delete-dataset help
rm -f part-991-request.json part-991-response.json part-991-payload.bin part-991-debug.log

Local request and response files may contain account IDs, ARNs, names, or service configuration. Remove them when the lab is complete and follow dependency-aware cleanup for any AWS resource you created.

Next, we will learn Rekognition: Create face liveness session from the CLI.

Official AWS CLI reference