Midterm - MCO (Building an AUF Solution with a Pre-Trained AI Model)

What is a pre-trained model?

a) A model currently being trained b) A model already trained by someone else that you download and use c) A model with no weights d) A model that must be retrained before use

b

What are the two required script tags for any page using a pre-trained TensorFlow.js model?

a) One for TensorFlow.js and one for the chosen model b) One for TensorFlow.js and one for jQuery c) One for the model and one for tfjs-vis d) One for TensorFlow.js and one for the CSS framework

a

Complete the script tag that loads TensorFlow.js: <script src="https://cdn.jsdelivr.net/npm/@tensorflow/\_\_\_\_"></script>

a) models b) tfjs c) core d) tf

b

COCO-SSD takes what as input, and what does it produce?

a) Text → sentiment scores b) Image/video → boxes + labels + confidence for ~80 object classes c) Audio → transcribed text d) Image/video → body keypoints

b

Approximately how many object classes can COCO-SSD detect?

a) 17 b) 21 c) 468 d) ~80

d

Which model produces a per-pixel person mask (foreground vs background)?

a) Body Segmentation (BodyPix/Selfie) b) Object Detection (COCO-SSD) c) Pose Estimation (MoveNet) d) Face Mesh

a

Which model takes image/video and produces 17 body keypoints with x, y, and score?

a) COCO-SSD b) MoveNet c) Face Mesh d) Hand Pose

b

What does the Text Toxicity model take as input and produce?

a) Text → a summary b) Text → a translation c) Text → probability scores across toxicity categories d) Text → 468 landmarks

c

What does the Speech Commands model take as input and produce?

a) Image → text b) Video → keypoints c) Text → audio d) Microphone audio → a recognized short spoken word/label

d

How many 3-D facial landmarks does Face Mesh produce?

a) 468 b) 17 c) 21 d) 80

a

How many hand keypoints per hand does Hand Pose produce?

a) 17 b) 21 c) 468 d) 80

b

What does the MobileBERT QnA model take as input and produce?

a) Passage only → a summary b) Question only → a generated answer c) Passage + question → the best answer span from the passage d) Audio + text → a transcript

c

Which model is described as a large download (~103 MB), where you should be patient on first load?

a) COCO-SSD b) MoveNet c) Face Mesh d) MobileBERT QnA

d

Which model would be most appropriate for virtual or blurred backgrounds in AUF online classes?

a) Body Segmentation (BodyPix/Selfie) b) COCO-SSD c) Text Toxicity d) Speech Commands

a

Which model would be most appropriate for an exercise-form checker in PE classes?

a) Face Mesh b) Pose Estimation (MoveNet) c) COCO-SSD d) Text Toxicity

b

Which model would be most appropriate for a hands-free "next/back" control for a presentation kiosk?

a) Hand Pose b) Face Mesh c) Speech Commands d) MoveNet

c

Which model would be most appropriate for a reading assistant that answers questions about a passage?

a) Text Toxicity b) COCO-SSD c) MobileBERT QnA d) Face Mesh

c

Which model would be most appropriate for touchless kiosk control using gestures?

a) Hand Pose (Hand Pose Detection) b) Speech Commands c) Text Toxicity d) MoveNet

a

Which model would be most appropriate for pre-screening posts in a student forum for harmful content?

a) Face Mesh b) Text Toxicity c) COCO-SSD d) Body Segmentation

b

Which model would be most appropriate for counting free seats in a library from a photo?

a) MoveNet b) Face Mesh c) Object Detection (COCO-SSD) d) Text Toxicity

c

Which model would be most appropriate for AR filters and frames for AUF events?

a) Hand Pose b) MoveNet c) COCO-SSD d) Face Mesh (Face Landmarks)

d

What environment must image/video and microphone models have to access the camera or microphone?

a) file:// b) HTTPS or localhost c) An FTP server d) A local file system

b

Why must the page be served with Live Server rather than opened directly?

a) Because Live Server compresses the files b) Because file:// paths cannot contain JavaScript c) Because opening the file directly will break model/asset loading d) Because the browser blocks all network access by default

c

When the instructor says "grade the use case as hard as the code," what does this mean?

a) Only code quality matters b) Only the use case matters c) The point is connecting a ready-made model to a real AUF need, not just getting a demo to run d) The demo does not need to run

c

Complete the second script tag so it loads the COCO-SSD model: <script src="https://cdn.jsdelivr.net/npm/@tensorflow-models/\_\_\_\_"></script>

a) coco-ssd b) mobilenet c) posenet d) speech-commands

a

What does the "start" action on the Speech Commands demo require?

a) Typing a word b) Uploading a file c) Clicking "Start" and saying the on-screen words d) Drawing on a canvas

c

Which demo feature should you watch when loading the QnA model on a slow connection?

a) The Elements tab b) The Console tab c) The Application tab d) The DevTools Network tab for "finished loading"

d

Using an AI assistant to help build the app is:

a) Allowed and encouraged, as long as you understand and can explain what you turned in b) Prohibited c) Allowed only for styling d) Allowed only if you cite the AI as an author

a

In the write-up, what must you do regarding the AI tool you used?

a) Name only the tool b) Describe only what it got wrong c) Name the tool and describe how you used it — what you prompted, what it got wrong, what you fixed d) Provide a transcript of the entire chat

c

What does the document compare the AI assistant to?

a) A replacement for the engineer b) A very fast junior partner — you are still the engineer c) A senior developer d) A code reviewer

b

Complete the idea of responsible AI tool use: if the app doesn't run or you can't explain it when asked, what is the result?

a) It gets a partial grade b) It counts anyway c) It doesn't count d) It gets extra credit for effort

c

Which model reads text and produces probability scores across toxicity categories?

a) MobileBERT QnA b) Text Toxicity c) Speech Commands d) Face Mesh

b

Which input type does the Body Segmentation model require?

a) Text b) Audio c) Image / video d) Numeric table

c

Which model produces a mask that distinguishes foreground from background?

a) Hand Pose b) Face Mesh c) Text Toxicity d) Body Segmentation (BodyPix/Selfie)

d

What is the general pattern that all eight models in the catalog follow?

a) They all read text and output text b) They all read an input (image, video, text, or audio) and produce a structured output c) They all require a GPU cluster d) They all need a Python backend

b

Which of the following is NOT one of the models in the catalog?

a) Object Detection (COCO-SSD) b) Pose Estimation (MoveNet) c) Sentiment Analysis (DistilBERT) d) Face Mesh (Face Landmarks)

c

Why does the instructor warn about the QnA model specifically?

a) Because it is a large download (~103 MB) b) Because it never finishes loading c) Because it uses the webcam d) Because it requires a special browser

a

What does "file:// does not satisfy the requirement" mean in this context?

a) Only file:// works for camera access b) file:// paths are too slow c) file:// paths do not allow camera/mic access, so Live Server (localhost) must be used d) file:// paths block TensorFlow.js entirely

c

The COCO-SSD model can detect roughly 80 object classes.

True

BodyPix/Selfie returns body keypoints rather than a person mask.

False

MoveNet produces 17 body keypoints per person.

True

Face Mesh produces 21 hand keypoints per hand.

False

Opening the HTML file directly with file:// still allows camera and microphone access.

False

The QnA model is a large download (~103 MB), so you should check the DevTools Network tab for "finished loading" before typing a question.

True

Using an AI assistant to help build the app is prohibited for this activity.

False

If your app doesn't run or you can't explain it when asked, it still counts toward your grade.

False

Speech Commands takes image/video input, while COCO-SSD takes microphone input.

False