This repo contains examples of capabilities of rekognition. Following examples are available in this repo:

• detecting custom labels
• detecting random labels
• detecting faces with bounding boxes
• detecting labels in videos

for detecting custom labels, the following pre-requisites should be met:

• create three s3 buckets for containing training data, testing data and output of rekognition training job.
• ensure than rekognition has appropriate permissions on those s3 buckets. for details on how to set this up, please refer to: https://docs.aws.amazon.com/rekognition/latest/customlabels-dg/su-sdk-bucket-permssions.html
• understand how the manifest file for custom label looks like. read: https://docs.aws.amazon.com/rekognition/latest/ customlabels-dg/cd-required-fields.html
• unzip the training_dataset.zip and upload the contents of it to the training bucket you created in first step.
• unzip the content of test_dataset.zip and upload the contents of it to the testing dataset bucket you created in first step.
• (Optional) change the region from us-east-1 to your choice of region, if you want to. you can find the region value in scripts/custom_labels/custom_label_runner.py

once all the above points are setup, run the custom_label_runner.py located under scripts/custom_labels. it will ask for a number of inputs like project name, version number, bucket locations etc. Once all the info is filled up correctly, it will start the job. The training job might take around 30-40 minutes to complete. Once the job is complete, just go back to the rekognition console, and you will see lot of information available regarding the model which has just been training along with an endpoint which you can deploy to start using it.

the same datasets are already present under scripts/custom_labels folder. you can extract it and upload the training and testing images in different s3 buckets with their manifest files respectively.