Exam AWS Certified Machine Learning - Specialty topic 1 question 49 discussion

Answer is "A". C and D are out as DeepLens is not offered as a commercial product. It is purely for developers to experiment with. From https://aws.amazon.com/deeplens/device-terms-of-use/ " (i) you may use the AWS DeepLens Device for personal, educational, evaluation, development, and testing purposes, and not to process your production workloads;" A is correct as it's will analyse live video streams instead of images. From https://aws.amazon.com/rekognition/video-features/ "Amazon Rekognition Video can identify known people in a video by searching against a private repository of face images. "

Why not A? DeepLens is for development purpose and much more expensive than just a camera. They are referring to 1000 camera in production scale?

Ans - A -- Proxy Server + Kinesis Video Streams + Rekognition Video The goal is to scale from 100 cameras to thousands and perform real-time detection of non-employees in office locations globally. The best approach is to use Amazon Kinesis Video Streams + Amazon Rekognition Video for real-time face detection.

The correct answer is D. Very tricky one but re-read the 2nd sentence in the question; “Images from the cameras were uploaded to Amazon S3 and tagged using Amazon Rekognition, and the results were stored in Amazon ES.” So, we have ‘images’ as training data, not videos. This is why it can not be option C - where it says to use Amazon Recognition Video. The only option mentioning Amazon Recognition Image is the option D. Also check: https://docs.aws.amazon.com/rekognition/latest/dg/what-is.html “...For example, each time a person arrives at your residence, your door camera can upload a photo of the visitor to Amazon S3. This triggers a Lambda function that uses Amazon Rekognition API operations to identify your guest. You can run analysis directly on images that are stored in Amazon S3 without having to load or move the data.”

A is answer

A not B: Use Amazon Rekognition Video instead of Amazon Rekognition Image in this case.

A is correct!

DeepLens is overkill for mass systems

A. NO - thousands of cameras would choke network bandwidth B. NO - thousands of cameras would choke network bandwidth C. YES - DeepLens is made for edge computing; it might be EOL / Not commercially available, but if they did not want you to use DeepLens the question would not have come in the first place D. NO - use Amazon Rekognition Video directly instead of Amazon Rekognition Image

Why A and not B? Can someone please explain it?

Option A

From Chat GPT The solution that the agency should consider is option A: Use a proxy server at each local office and for each camera, and stream the RTSP feed to a unique Amazon Kinesis Video Streams video stream. On each stream, use Amazon Rekognition Video and create a stream processor to detect faces from a collection of known employees and alert when non-employees are detected. By using a proxy server at each local office and streaming the RTSP feed to individual Amazon Kinesis Video Streams video streams, the agency can efficiently handle the large number of video cameras in different office locations. Using Amazon Rekognition Video, the agency can create a stream processor to detect faces from a collection of known employees. This allows for real-time identification of non-employees based on facial recognition. Alerts can then be generated when non-employees are detected, ensuring that the agency is able to identify and respond to potential security threats in real-time.

I initially thought it is C but looks like A makes more sense here.

The DeepLens Service will reach EOL at the end of Jan 2024, so more than likely that this question will not be asked in the exam

D is the answer now, DeepLens is used for situations like this!

Think big picture - you tested something (let say code python) and ready to implement into prod will you move python code or java code! Here in this particular case, they tested with actual video camera and they did not say deeplense so answer is A! For knowledge sake if they say in real exam it is tested with deeplense ---then ideal solution should be model inference happening at deeplense itself with search against existing employees and send back model inference when it detect new faces who are not employees back to cloud may be S3

Answer is "A". As mentioned in below user comment, DeepLens is not offered as a commercial product. https://aws.amazon.com/deeplens/device-terms-of-use/