Question 300
Scanned shipping labels are legible to humans but a caption endpoint outputs “parcel on table.” What feature should be requested?
- Request dense image captions or another region-aware description capability when available
- Use OCR or Read text extraction and consume recognized lines and words
- Consume separate detected object instances with their bounding regions
- Use supported smart-crop region suggestions and apply them to thumbnail generation
Explanation
Correct Answer
B:Scene captioning summarizes an image; OCR targets the exact characters on printed labels. In the e-commerce image processing service, confirm the fix by retesting the affected behavior after using OCR or Read text extraction and consume recognized lines and words.
Incorrect Answers
A:That change addresses another failure mode: a single global caption can omit local items; dense or region-level descriptions convey additional spatially localized visual content; it does not address this case, for which the relevant intervention is to use OCR or Read text extraction and consume recognized lines and words.
C:Per-object detections support counting and localization; one image-level class does not distinguish instances in its own workflow, but that is a different correction from the present requirement to use OCR or Read text extraction and consume recognized lines and words. The reported issue would remain.
D:Although useful elsewhere in the e-commerce image processing service, smart crop identifies an informative image region; it does not directly alter the original product catalog metadata; the present case is different and calls for the team to use OCR or Read text extraction and consume recognized lines and words.
Question 301
In a factory quality-control camera, engineers must let a manufacturing application send new images to the trained model. The expected evidence is that the client obtains class predictions from the published iteration. Which implementation is the best match?
- Train a multilabel image classifier
- Publish and consume a trained prediction endpoint
- Train a custom object-detection model
- Train a multiclass image classifier
Explanation
Correct Answer
B:Publishing exposes a trained iteration through a supported prediction interface requiring appropriate credentials. For this factory quality-control camera, the intended evidence is that the client obtains class predictions from the published iteration. Training a model does not automatically connect the inference call from a production system.
Incorrect Answers
A:Multilabel classification allows multiple class labels to apply simultaneously rather than forcing one exclusive choice; it may be used to identify several independent fault categories present in the same image in a factory quality-control camera, but the current requirement is instead to let a manufacturing application send new images to the trained model. That separate capability does not prove that the client obtains class predictions from the published iteration; It does not inherently produce a bounding box for each labeled fault, so selecting that alternative would not meet the stated acceptance check.
C:Custom object detection uses labeled regions to learn both class prediction and object location; it may be used to recognize a company-specific defect and identify where it appears in a factory quality-control camera, but the current requirement is instead to let a manufacturing application send new images to the trained model. That separate capability does not prove that the client obtains class predictions from the published iteration; Whole-image classification lacks the required per-defect coordinates, so selecting that alternative would not meet the stated acceptance check.
D:Multiclass classification predicts a single label from the project's mutually exclusive classes; it may be used to assign exactly one defined quality category to each whole image in a factory quality-control camera, but the current requirement is instead to let a manufacturing application send new images to the trained model. That separate capability does not prove that the client obtains class predictions from the published iteration; It does not identify the coordinates of individual defects in the photograph, so selecting that alternative would not meet the stated acceptance check.
Question 302
A factory image always shows exactly one dominant defect class, and every training image has one label. Which labeling arrangement fits?
- Draw bounding boxes around each relevant defect instance
- Review class balance and per-class precision and recall on representative samples
- Use multilabel annotations allowing several positive classes on one image
- Assign one mutually exclusive class label to each image for multiclass training
Explanation
Correct Answer
D:Multiclass classification predicts one category among alternatives for an entire image. In the factory quality-control camera, confirm the fix by retesting the affected behavior after assigning one mutually exclusive class label to each image for multiclass training.
Incorrect Answers
A:Although useful elsewhere in the factory quality-control camera, object detection training requires localized annotated instances, not only image-level class labels; the present case is different and calls for the team to assign one mutually exclusive class label to each image for multiclass training.
B:An imbalanced dataset can yield high overall accuracy while missing the rare important defects in its own workflow, but that is a different correction from the present requirement to assign one mutually exclusive class label to each image for multiclass training. The reported issue would remain.
C:That change addresses another failure mode: multilabel classification models independent class presence rather than enforcing a single winner; it does not address this case, for which the relevant intervention is to assign one mutually exclusive class label to each image for multiclass training.
Question 303
A labeled dataset contains 95% good products and very few damaged ones. What should the team evaluate before deployment?
- Use multilabel annotations allowing several positive classes on one image
- Publish the intended model iteration to a prediction resource or supported deployment
- Assign one mutually exclusive class label to each image for multiclass training
- Review class balance and per-class precision and recall on representative samples
Explanation
Correct Answer
D:An imbalanced dataset can yield high overall accuracy while missing the rare important defects. In the factory quality-control camera, confirm the fix by retesting the affected behavior after reviewing class balance and per-class precision and recall on representative samples.
Incorrect Answers
A:Multilabel classification models independent class presence rather than enforcing a single winner in its own workflow, but that is a different correction from the present requirement to review class balance and per-class precision and recall on representative samples. The reported issue would remain.
B:That change addresses another failure mode: training output is not automatically a callable production prediction endpoint; it does not address this case, for which the relevant intervention is to review class balance and per-class precision and recall on representative samples.
C:Although useful elsewhere in the factory quality-control camera, multiclass classification predicts one category among alternatives for an entire image; the present case is different and calls for the team to review class balance and per-class precision and recall on representative samples.
Question 304
A factory quality-control camera must assign exactly one defined quality category to each whole image. Reviewers will accept the implementation when each input image has one predicted class and confidence. Which approach best fulfills that requirement?
- Train a multilabel image classifier
- Prepare and label representative training images
- Train a custom object-detection model
- Train a multiclass image classifier
Explanation
Correct Answer
D:Multiclass classification predicts a single label from the project's mutually exclusive classes. For this factory quality-control camera, the intended evidence is that each input image has one predicted class and confidence. It does not identify the coordinates of individual defects in the photograph.
Incorrect Answers
A:Multilabel classification allows multiple class labels to apply simultaneously rather than forcing one exclusive choice; it may be used to identify several independent fault categories present in the same image in a factory quality-control camera, but the current requirement is instead to assign exactly one defined quality category to each whole image. That separate capability does not prove that each input image has one predicted class and confidence; It does not inherently produce a bounding box for each labeled fault, so selecting that alternative would not meet the stated acceptance check.
B:High-quality labels and representative images establish the supervised training signal for a custom vision project; it may be used to provide the custom model with accurately annotated examples of the target classes in a factory quality-control camera, but the current requirement is instead to assign exactly one defined quality category to each whole image. That separate capability does not prove that each input image has one predicted class and confidence; Publishing an empty project or adding only unlabeled production images will not define the target behavior, so selecting that alternative would not meet the stated acceptance check.
C:Custom object detection uses labeled regions to learn both class prediction and object location; it may be used to recognize a company-specific defect and identify where it appears in a factory quality-control camera, but the current requirement is instead to assign exactly one defined quality category to each whole image. That separate capability does not prove that each input image has one predicted class and confidence; Whole-image classification lacks the required per-defect coordinates, so selecting that alternative would not meet the stated acceptance check.
Question 305
Photos from the same manufactured unit appear in both training and test sets. What problem does that create?
- Collect representative images from the new camera setup and reevaluate or retrain
- Collect labeled examples of every supported target class and retrain
- Separate correlated units across train and evaluation splits
- Call the prediction service through a secured backend or identity-aware gateway
Explanation
Correct Answer
C:Near-identical images in both sets leak information and exaggerate estimated generalization. In the factory quality-control camera, confirm the fix by retesting the affected behavior after separating correlated units across train and evaluation splits.
Incorrect Answers
A:That change addresses another failure mode: distribution shifts in scale, lighting, and viewpoint can degrade custom detectors; it does not address this case, for which the relevant intervention is to separate correlated units across train and evaluation splits.
B:A supervised closed-set classifier cannot reliably learn a class that it was never taught in its own workflow, but that is a different correction from the present requirement to separate correlated units across train and evaluation splits. The reported issue would remain.
D:Although useful elsewhere in the factory quality-control camera, exposing service credentials in a public client undermines key secrecy and access governance; the present case is different and calls for the team to separate correlated units across train and evaluation splits.
Question 306
Inspectors need the precise location of a chipped component rather than only the image's overall category. What labels should training examples contain?
- Draw bounding boxes around each relevant defect instance
- Review class balance and per-class precision and recall on representative samples
- Assign one mutually exclusive class label to each image for multiclass training
- Publish the intended model iteration to a prediction resource or supported deployment
Explanation
Correct Answer
A:Object detection training requires localized annotated instances, not only image-level class labels. In the factory quality-control camera, confirm the fix by retesting the affected behavior after drawing bounding boxes around each relevant defect instance.
Incorrect Answers
B:An imbalanced dataset can yield high overall accuracy while missing the rare important defects in its own workflow, but that is a different correction from the present requirement to draw bounding boxes around each relevant defect instance. The reported issue would remain.
C:Although useful elsewhere in the factory quality-control camera, multiclass classification predicts one category among alternatives for an entire image; the present case is different and calls for the team to draw bounding boxes around each relevant defect instance.
D:That change addresses another failure mode: training output is not automatically a callable production prediction endpoint; it does not address this case, for which the relevant intervention is to draw bounding boxes around each relevant defect instance.
Question 307
A multilabel predictor returns a low-confidence label the UI shows as confirmed. What should the client configure?
- Standardize the annotation taxonomy before training new versions
- Review bounding-box annotation quality and small-object representation
- Verify the model iteration and prediction endpoint referenced by the client
- Evaluate class-specific confidence thresholds against validation data
Explanation
Correct Answer
D:Multiple class scores need thresholding; accepting every nonzero prediction creates false positives. In the factory quality-control camera, confirm the fix by retesting the affected behavior after evaluating class-specific confidence thresholds against validation data.
Incorrect Answers
A:Although useful elsewhere in the factory quality-control camera, inconsistent labels teach contradictory distinctions and reduce model reliability; the present case is different and calls for the team to evaluate class-specific confidence thresholds against validation data.
B:That change addresses another failure mode: misaligned or sparse localization labels compromise box prediction despite correct class vocabulary; it does not address this case, for which the relevant intervention is to evaluate class-specific confidence thresholds against validation data.
C:A retrained iteration has no effect when clients continue invoking the old published version in its own workflow, but that is a different correction from the present requirement to evaluate class-specific confidence thresholds against validation data. The reported issue would remain.
Question 308
An app receives predictions from the previous model despite new training completing. What should be inspected?
- Verify the model iteration and prediction endpoint referenced by the client
- Evaluate class-specific confidence thresholds against validation data
- Review bounding-box annotation quality and small-object representation
- Standardize the annotation taxonomy before training new versions
Explanation
Correct Answer
A:A retrained iteration has no effect when clients continue invoking the old published version. In the factory quality-control camera, confirm the fix by retesting the affected behavior after verifying the model iteration and prediction endpoint referenced by the client.
Incorrect Answers
B:That change addresses another failure mode: multiple class scores need thresholding; accepting every nonzero prediction creates false positives; it does not address this case, for which the relevant intervention is to verify the model iteration and prediction endpoint referenced by the client.
C:Misaligned or sparse localization labels compromise box prediction despite correct class vocabulary in its own workflow, but that is a different correction from the present requirement to verify the model iteration and prediction endpoint referenced by the client. The reported issue would remain.
D:Although useful elsewhere in the factory quality-control camera, inconsistent labels teach contradictory distinctions and reduce model reliability; the present case is different and calls for the team to verify the model iteration and prediction endpoint referenced by the client.
Question 309
A production class never appeared in the training set but is expected at inference. What must be corrected?
- Call the prediction service through a secured backend or identity-aware gateway
- Preserve each valid class prediction independently instead of taking only the top class
- Collect representative images from the new camera setup and reevaluate or retrain
- Collect labeled examples of every supported target class and retrain
Explanation
Correct Answer
D:A supervised closed-set classifier cannot reliably learn a class that it was never taught. In the factory quality-control camera, confirm the fix by retesting the affected behavior after collecting labeled examples of every supported target class and retrain.
Incorrect Answers
A:Although useful elsewhere in the factory quality-control camera, exposing service credentials in a public client undermines key secrecy and access governance; the present case is different and calls for the team to collect labeled examples of every supported target class and retrain.
B:That change addresses another failure mode: an argmax conversion defeats the purpose of multilabel output when several classes may apply; it does not address this case, for which the relevant intervention is to collect labeled examples of every supported target class and retrain.
C:Distribution shifts in scale, lighting, and viewpoint can degrade custom detectors in its own workflow, but that is a different correction from the present requirement to collect labeled examples of every supported target class and retrain. The reported issue would remain.
Question 310
A custom vision model was trained, but prediction requests target an unpublished iteration. What release step is missing?
- Assign one mutually exclusive class label to each image for multiclass training
- Review class balance and per-class precision and recall on representative samples
- Use multilabel annotations allowing several positive classes on one image
- Publish the intended model iteration to a prediction resource or supported deployment
Explanation
Correct Answer
D:Training output is not automatically a callable production prediction endpoint. In the factory quality-control camera, confirm the fix by retesting the affected behavior after publishing the intended model iteration to a prediction resource or supported deployment.
Incorrect Answers
A:Multiclass classification predicts one category among alternatives for an entire image in its own workflow, but that is a different correction from the present requirement to publish the intended model iteration to a prediction resource or supported deployment. The reported issue would remain.
B:Although useful elsewhere in the factory quality-control camera, an imbalanced dataset can yield high overall accuracy while missing the rare important defects; the present case is different and calls for the team to publish the intended model iteration to a prediction resource or supported deployment.
C:That change addresses another failure mode: multilabel classification models independent class presence rather than enforcing a single winner; it does not address this case, for which the relevant intervention is to publish the intended model iteration to a prediction resource or supported deployment.