Exam AWS Certified Machine Learning - Specialty topic 1 question 167 discussion

A : The data has label. So what we need to do is to enforce accuracy by reviewing low confidence ones internally

D; B is using MTurk which uses public workforce which violates the requirements that videos need to be kept private

- A2I: Focuses on integrating human review into the decision-making process of models post-prediction. - Ground Truth: Focuses on creating high-quality labeled datasets during the training phase.

"However, the object detection model cannot clearly demarcate the yellow line, the passengers who cross the yellow line, and the trains." ---> D

IMHO, the answer is D: image segmentation. As the question say: "However, the object detection model cannot clearly demarcate the yellow line, the passengers who cross the yellow line, and the trains." You will get a better accuracy with segmentation in this case.

semantic segmentation may not be the right choice for labelling, instead Rekognition is ideal for this scenario and private workforce + A2I to validate the labelling.

While Amazon Rekognition Custom Labels with Amazon A2I could be used for object detection, semantic segmentation provides more detailed information about the spatial layout of objects in an image, making it potentially more suitable for tasks like demarcating safety lines.

Semantic segmentation will provide the precise pixel-level labeling required to demarcate the yellow safety line accurately, passengers, and trains. A private workforce will ensure that the video data remains confidential. As the original model can’t correctly identify the line, semantic segmentation might offer the needed precision. So D is right.

Amazon Rekognition does not support creation of private workforce. Between A & D, D is the only option that allows its creation. Semantic segmnentation can easily identify the yellow line.

Given the requirements of the task and the need for confidentiality, the best approach would be: D. Use an Amazon SageMaker Ground Truth semantic segmentation labeling task with a private workforce. Semantic segmentation will provide the precise pixel-level labeling required to demarcate the yellow safety line accurately, passengers, and trains. A private workforce will ensure that the video data remains confidential.

D. Use an Amazon SageMaker Ground Truth semantic segmentation labeling task. Use a private workforce as the labeling workforce. Here's why this approach is suitable: Semantic Segmentation Labeling: Semantic segmentation involves labeling each pixel in the image, which is more granular than bounding boxes. This approach is ideal for accurately demarcating the yellow line, which might be difficult with just bounding boxes. It also allows for precise detection of passengers and trains. Private Workforce: Given the requirement for confidentiality, using a private workforce ensures that the data is handled by trusted, authorized personnel. This addresses the concern of keeping the video data confidential. Amazon SageMaker Ground Truth: This service provides tools for efficient and accurate labeling of image data, which is essential for training a robust object detection model.

A: D is too complicated. Option D suggests using SageMaker Ground Truth with a semantic segmentation labeling task and a private workforce. Semantic segmentation can be useful for delineating the yellow line clearly. However, it might be more complex than necessary for this scenario, and object detection might be more suitable. In my opinion, A is a better option.

Rekognition is not guaranteed no to use your data to improve their models. Similarly, mechanical turk will not keep data private. Only viable option is D

B and C excluded since use public workforce. D excluded since question is asking for "labeling approach for THIS model" it means you don't want to switch to a semantic segmentation problem. Therefore A is the correct answer

Given that the video data must remain confidential, options that use public workforces like Amazon Mechanical Turk or third-party AWS Marketplace vendors would not be suitable. option A relies on object detection (bounding boxes) and doesn't switch to semantic segmentation. semantic segmentation provides precise labels, especially when distinguishing between closely placed objects like the yellow line and the passengers.

A: Using Amazon Rekognition Custom Labels you can do the same of Ground Truth this answer is complete with all steps.

The company can use Amazon Rekognition Custom Labels to label the dataset and create a custom Amazon Rekognition object detection model. They can create a private workforce and use Amazon Augmented AI (Amazon A2I) to review the low-confidence predictions and retrain the custom Amazon Rekognition model. This approach will help the company improve the model as it allows them to train a custom model that is specific to their business needs. The custom model can be trained to detect the yellow line, passengers who cross the yellow line, and trains in the video feeds. The private workforce ensures that the video data remains confidential, while Amazon A2I helps to improve the accuracy of the model by reviewing low-confidence predictions and retraining the model. which make A is more suitable than D using the Sagemaker Ground Truth semantic segmentation.