The plugin is installed by dragging ‘AmazonAI.catdv’ onto the Worker Node or placing it in the Worker Extensions folder. Once installed, four new actions become available in the Worker Node:
Amazon Image Rekognition
Amazon Video Rekognition
How it Works
The actions which use media (i.e. Rekognition and Transcribe) will:
encode an AI-optimized temp file (or use the original video where appropriate)
upload it to your remote S3 volume
feed that location to the appropriate AI service
collect and populate the results either as text/html in a field of your choice, or clip markers for transcribe and video tagging operations.
delete the temp file from the remote S3 volume
Image Rekognition and Translate actions run ‘on the fly’, meaning the processing results are returned in a few seconds and there is no need to retrieve results later. However, Transcribe and Video Rekognition actions operate in an ‘asynchronous’ context, meaning that jobs must be submitted and retrieved by separate worker actions. For asynchronous jobs (Transcribe and Video Rekognition) the results are saved on AWS for 7 days - meaning you can re-submit your retrieval action up to 7 days later and rather than re-running the job at additional cost, CatDV will reprocess the existing results. The Worker action would fail if you submit a retrieval for a job older than 7 days.
How these actions are defined depends on your specific workflow needs, but a basic example would be to submit a transcribe job using a checkbox or pick list value, with the retrieval job running on a 2-minute cadence meaning it likely only needs to check once or twice before the job results are returned. For job retrieval actions you could monitor the “aws.ai.rekognition.job.status” field for a value of “IN_PROGRESS”, meaning the job is still active. See ‘Troubleshooting & Advanced Notes’ for more details on the fields used by the plugin.
All actions require a set of IAM access keys, however for Rekognition/Transcribe which both utilize an S3 remote volume you are only required to confirm the Secret Key. If you wish to use a different account for the Rekognition/Transcribe processing versus the S3 storage, a separate access key can optionally be entered, with the only requirement being that the region matches that of the remote volume: