Skip to content
By using AWS re:Post, you agree to the AWS re:Post Terms of Use
AWS re:Postre:Post
About re:Post

AWS Rekognition Limits

0

Hey Guys!

I'm designing a solution that requires real-time video analysis using Amazon Rekognition. Based on my research, I understand there are certain limits, but I need clarification on several aspects:

  • For stored video: Maximum of 5 TPS or 20 concurrent jobs per account?
  • For Rekognition Streaming: 5 transactions per second limit?
  • What mean that each Kinesis Video stream can only be associated with one stream processor? Does that impact on something?

So, reading the docs,

  1. What is the maximum number of concurrent video streams that can be analyzed in real-time with Rekognition? Is the limit of 5 TPS for Image API applicable to video frame analysis as well?

  2. Can Custom Labels be used with streaming video analysis? If so, what are the limits for concurrent processing?

  3. Can Service Quotas increases be requested for video processing specifically? Are there different quotas for different video analysis operations (label detection vs. content moderation vs. Custom labels)?

Finally, For a system that needs to monitor 20+ concurrent video streams for brand safety/content moderation, what would be the recommended approach to handle this volume while staying within service limits?

Any insights from those who have scaled Rekognition video processing would be greatly appreciated!

Topics
AnalyticsMachine Learning & AIMedia Services
Tags
Service QuotasAmazon RekognitionLive Video StreamingVideo-on-Demand StreamingImage Processing & Analytics
Language
English
asked a year ago504 views
1 Answer
0

Thank you for your questions about Amazon Rekognition's limits and capabilities. I'll address your points based on the available information:

  1. For real-time video analysis with Amazon Rekognition Streaming Video Events, the service can support up to 600 concurrent sessions per AWS customer. This is significantly higher than the 20 concurrent jobs limit for stored video analysis. The 5 TPS (Transactions Per Second) limit you mentioned is not directly applicable to streaming video analysis. Each stream is processed continuously rather than as discrete transactions.

  2. Custom Labels are not currently mentioned as being available for use with streaming video analysis. Custom Labels are typically used with image analysis and stored video analysis, but not with real-time streaming video.

  3. Service Quotas increases can indeed be requested for video processing. Different video analysis operations may have different quotas. For example, there are specific quotas for Face Liveness API operations in different regions. However, the exact quotas for each type of video analysis operation (label detection, content moderation, etc.) are not explicitly stated in the provided information. It's best to check the AWS Service Quotas console for the most up-to-date information on specific limits and whether they can be increased.

Regarding your system that needs to monitor 20+ concurrent video streams for brand safety/content moderation, here are some recommendations:

  1. Since Amazon Rekognition Streaming Video Events can support up to 600 concurrent sessions, you should be able to handle 20+ streams within the default limits.

  2. Be aware that each Kinesis Video stream can only be associated with one stream processor. This means you'll need to set up a separate stream processor for each video stream you want to analyze.

  3. Consider using the region of interest feature when configuring your StreamProcessor. This allows you to choose a specific area of the frame to process, which can be more efficient if you don't need to analyze the entire frame.

  4. If you need to scale beyond the default limits, reach out to your AWS account manager to discuss increasing your concurrent session limit.

  5. Implement a robust error handling and retry mechanism in your application to manage any potential service limits or throttling issues.

  6. Monitor your usage closely and set up alerts to notify you when you're approaching service limits.

Remember to thoroughly test your system to ensure it can handle the required volume of streams while staying within service limits. If you find you need higher limits, don't hesitate to contact AWS support for assistance.
Sources
Amazon Rekognition – frequently asked questions - AWS
Amazon Rekognition endpoints and quotas - AWS General Reference
Guidelines and quotas in Amazon Rekognition - Amazon Rekognition

answered a year ago
EXPERT
reviewed a year ago
  • Javier Flores
    a year ago

    With the 20 concurrent job limit for stored video analysis, what is the recommended approach if we need to process more than 20 videos simultaneously during peak periods? Watch length should the video be?

    And what is the maximum video duration supported for a single batch processing job? Will I receive live notifications of violations during that period?

    Does AWS offer reserved capacity options for Rekognition batch processing to reduce costs for predictable workloads?

    What is the process and typical timeline for requesting quota increases for concurrent batch video jobs?

    For content moderation specifically, at what granularity does Rekognition provide timestamps for detected violations?

    Is there a recommended sampling rate for frame analysis to balance thoroughness with processing costs?

    What is the recommended architecture for implementing a queuing system to manage video processing within the 20 concurrent job limit?

    How can we effectively implement a notification system for completed analysis that scales with hundreds of videos per hour?

    What is the typical end-to-end latency we should expect from job submission to completion for a 10-minute video or more?

    Can the results of batch video analysis be used to train custom label models for future detection improvements?

    What do we need to train the model to detect custom levels for videos?

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.

Relevant content