MCQs

Disclaimer:

Educational Use Only:

The website links provided are to be used strictly for educational and knowledge-gaining purposes only.

No Misuse:

Do not misuse the links or the information contained therein for any commercial, illegal, or unethical activities.

Respect for Websites and Businesses:

Respect the intellectual property and business interests of the websites and their owners. Do not attempt to alter, copy, or redistribute content without proper authorization.

Acknowledge Value of Knowledge:

Recognize and value the knowledge and resources shared by these websites. Use the information responsibly and ethically.

No Endorsement:

The provision of these links does not imply endorsement of the linked sites, their content, or their products and services. Users are responsible for evaluating the suitability and accuracy of the information.

User Responsibility:

By accessing these links, you agree to use the information at your own risk and to comply with all applicable laws and website terms of use.

Overall MCQs for Reference

What is the main advantage of the Single Shot MultiBox Detector (SSD) model?
a. Higher resolution outputs
b. Simpler anchor box selection
c. Real-time object detection speed
d. Reduces the number of object proposals needed
Answer: c. Real-time object detection speedslideshare+1
The SSD model primarily uses which base network for feature extraction in its original version?
a. ResNet-50
b. AlexNet
c. VGG16
d. Inception-v3
Answer: c. VGG16unc+1
Which of the following is NOT true about SSD?
a. Generates object proposals
b. Uses multiple feature maps
c. Handles objects of various sizes
d. Performs detection in a single pass
Answer: a. Generates object proposalsd2l+1
In SSD, what is a “default box”?
a. The ground-truth bounding box
b. Predefined bounding box with various aspect ratios
c. Randomly placed bounding box
d. Bounding box generated after NMS
Answer: b. Predefined bounding box with various aspect ratiosgeeksforgeeks+1
What is the main loss function used in SSD training?
a. Hinge loss
b. Cross-entropy loss
c. Combination of localization and confidence loss
d. Dice loss
Answer: c. Combination of localization and confidence lossarxiv+1
What does Non-Maximum Suppression (NMS) do in SSD?
a. Normalizes feature maps
b. Removes overlapping detections
c. Increases inference speed
d. Reduces input image size
Answer: b. Removes overlapping detectionsgeeksforgeeks+1
SSD can detect objects at multiple scales due to:
a. Multiple fully-connected layers
b. Global average pooling
c. Feature maps from different layers
d. Batch normalization
Answer: c. Feature maps from different layersclassic.d2l+1
The “confidence score” in SSD refers to:
a. IoU of anchor boxes
b. Probability of background class only
c. Probability of an object class per box
d. Loss penalization factor
Answer: c. Probability of an object class per boxgeeksforgeeks
Which step is performed FIRST in the SSD detection pipeline?
a. Applying NMS
b. Generating default boxes
c. Predicting bounding boxes and class scores
d. Filtering low-confidence detections
Answer: b. Generating default boxesclassic.d2l+1
What operation is commonly used for predictions in SSD?
a. 1×1 convolution
b. 3×3 convolution
c. Max pooling
d. Fully connected layer
Answer: b. 3×3 convolutionyoutubeclassic.d2l
For small object detection, SSD relies on:
a. Early high-resolution feature maps
b. Data dropout
c. Global feature maps
d. Only the deepest layer
Answer: a. Early high-resolution feature mapsslideshareyoutube
What is “hard negative mining” in SSD?
a. Removing redundant anchors
b. Balancing positive and negative samples in training
c. Focusing on hardest positive samples
d. Delaying optimization
Answer: b. Balancing positive and negative samples in trainingslideshare
Which of these datasets is renowned for SSD benchmarking?
a. CIFAR-10
b. PASCAL VOC
c. Kaggle Dogs vs Cats
d. MNIST
Answer: b. PASCAL VOCarxiv+1
SSD is compared to which two-stage detection model for speed?
a. YOLOv3
b. Mask R-CNN
c. Faster R-CNN
d. DenseNet
Answer: c. Faster R-CNNarxiv+1
In SSD, what helps in classification and precise localization simultaneously?
a. Dual networks
b. Anchor box refinements
c. Multi-task loss
d. Batch normalization
Answer: c. Multi-task lossgeeksforgeeks+1
The SSD detection framework is considered:
a. End-to-end trainable
b. Only for transfer learning
c. Unsupervised
d. Not suitable for mobile
Answer: a. End-to-end trainablearxiv
How does SSD handle class imbalance?
a. Uses weighted loss
b. Hard negative mining
c. Oversampling positive samples
d. Excluding background class
Answer: b. Hard negative miningslideshare
For each candidate box, SSD predicts:
a. Only objectness score
b. Class probabilities and bounding box offsets
c. Only bounding box offsets
d. Only class labels
Answer: b. Class probabilities and bounding box offsetsd2l+1
What kind of architecture is used to generate predictions at multiple scales in SSD?
a. Feature pyramid
b. Sequential pooling
c. Recurrent blocks
d. Skip connections
Answer: a. Feature pyramidclassic.d2l
Which is a key reason for SSD’s efficiency compared to older detector models?
a. Processing the image at multiple scales with a single forward pass
b. Only using hand-crafted features
c. Training for epochs
d. Relying on SVMs for classification
Answer: a. Processing the image at multiple scales with a single forward passslideshare+1
The main limitation of early SSD models is:
a. Poor accuracy on large objects
b. Poor speed for real-time tasks
c. Difficulty detecting small objects
d. Inflexible architecture
Answer: c. Difficulty detecting small objectsyoutubeslideshare
SSD applies which type of prediction to each default box?
a. SVM-based detection
b. Softmax-based classification
c. Max pooling localization
d. Logistic regression
Answer: b. Softmax-based classificationd2l+1
SSD generates how many sets of predictions at every location in a feature map?
a. 1
b. 2
c. As many as there are default boxes
d. Only for the highest scoring class
Answer: c. As many as there are default boxesclassic.d2l+1
The abbreviation “mAP” stands for:
a. multi-scale Accuracy Precision
b. mean Average Precision
c. multi-label Average Prediction
d. maximum Anchoring Precision
Answer: b. mean Average Precisionarxiv+1
During training, how are positive matches defined in SSD?
a. All anchors that overlap with ground truth
b. Anchors with IoU above a set threshold
c. Anchors with minimum aspect ratio
d. All anchors in highest confidence feature map
Answer: b. Anchors with IoU above a set thresholdslideshare+1
In SSD, what is the role of extra convolutional layers beyond the base network?
a. Increase feature map size
b. Enable generative capabilities
c. Detect larger objects and enable multiscale detection
d. Reduce model parameter count
Answer: c. Detect larger objects and enable multiscale detectiongeeksforgeeks+1
Which of these is an application area for SSD?
a. Time series forecasting
b. Audio denoising
c. Real-time video object detection
d. Text classification
Answer: c. Real-time video object detectionexamtopics+1
What strategy helps SSD maintain high inference speed?
a. Parallel processing with batch normalization
b. Single-pass computation without region proposals
c. Recurrent network-based proposals
d. Frame-wise sequential detection
Answer: b. Single-pass computation without region proposalsarxiv
In SSD, what characterizes an “anchor box”?
a. Fixed center and size
b. Multiple aspect ratios and scales
c. Only applied to deepest layers
d. One per pixel
Answer: b. Multiple aspect ratios and scalesclassic.d2l+1
The default size of input images for SSD300 is:
a. 224×224
b. 300×300
c. 416×416
d. 448×448
Answer: b. 300×300slideshare+1
The backbone network in SSD is typically pre-trained on which dataset?
a. COCO
b. VOC
c. ImageNet
d. CIFAR-100
Answer: c. ImageNetarxiv+1
Smooth L1 loss in SSD is primarily used for:
a. Class prediction
b. Localization error
c. Increasing recall
d. Reducing NMS
Answer: b. Localization errorgeeksforgeeks+1
Which approach has recently improved SSD small object detection?
a. Larger batch sizes
b. Additional context layers and feature fusion
c. Manual labeling
d. Discarding low-resolution features
Answer: b. Additional context layers and feature fusionclassic.d2l+1
SSD’s improvement over YOLO primarily lies in:
a. Use of max pooling
b. Multi-scale feature prediction
c. Use of R-NN region proposals
d. SVM classifiers
Answer: b. Multi-scale feature predictiond2l+1
What data augmentation strategy enhances SSD training?
a. Only horizontal flipping
b. Random cropping, photometric distortions, and zoom
c. Data standardization
d. Noise injection
Answer: b. Random cropping, photometric distortions, and zoomslideshare+1
SSD handles detection of different aspect ratios by:
a. Varying strides in convolutions
b. Using different scales and aspect ratios for default boxes
c. Increasing depth of base network
d. Reducing anchor count
Answer: b. Using different scales and aspect ratios for default boxesgeeksforgeeks+1
Non-maximum suppression (NMS) ensures:
a. Only the worst boxes are kept
b. Output is postprocessed for clarity
c. Only one bounding box per detected object is retained
d. Increased model complexity
Answer: c. Only one bounding box per detected object is retainedunc+1
On the PASCAL VOC 2007 dataset, what is the approximate mAP reported for SSD300?
a. 85%
b. 74.3%
c. 61.4%
d. 50.2%
Answer: b. 74.3%unc+1
The SSD algorithm is best described as:
a. Proposal-based
b. Segmentation-driven
c. Single-pass, proposal-free
d. Ensemble of binary classifiers
Answer: c. Single-pass, proposal-freed2l+1
For post-processing in SSD, which metric is crucial?
a. Precision-Recall Curve
b. Intersection over Union (IoU)
c. Mean Squared Error
d. Average Loss
Answer: b. Intersection over Union (IoU)unc+1
What is the typical stride for SSD feature map convolutional layers in the backbone?
a. 16
b. 8
c. Varies depending on the layer
d. 2
Answer: c. Varies depending on the layerclassic.d2l+1
In SSD training, which negative samples are prioritized for loss calculation?
a. Random negatives
b. Those with the lowest predicted confidence
c. Hard negatives with high confidence
d. Only positives are used
Answer: c. Hard negatives with high confidenceslideshare
The final classifier head in SSD produces:
a. A single class score for the whole image
b. Multiple class scores and box offsets per anchor
c. Segmentation masks
d. Only objectness metric
Answer: b. Multiple class scores and box offsets per anchorgeeksforgeeks+1
What does SSD eliminate compared to two-stage detectors?
a. Label encoding
b. Proposal generation and resampling
c. Feature extraction
d. NMS post-processing
Answer: b. Proposal generation and resamplingd2l+1
The key contribution of SSD in detection pipelines is:
a. Real-time trainability
b. Simultaneous localization and classification
c. Pixel-wise segmentation
d. Separate feature extraction
Answer: b. Simultaneous localization and classificationarxiv+1
Which of the following can be a limitation for deploying SSD on edge devices?
a. High parameter count with large base networks
b. Susceptibility to vanishing gradients
c. Lack of NMS
d. Requirement of region proposals
Answer: a. High parameter count with large base networksunc+1
SSD300 and SSD512 refer to:
a. Number of feature maps used
b. Input image resolution
c. Number of layers in model
d. Output classes
Answer: b. Input image resolutionarxiv+1
Which feature enables SSD to be easily adapted for new object categories?
a. Modular classifier and regression heads
b. Fixed number of detections
c. Use of handcrafted features
d. No need for retraining
Answer: a. Modular classifier and regression headsarxiv
In the SSD model, which part is responsible for predicting class scores and box positions?
a. Base network only
b. Extra convolutional predictor layers
c. NMS block
d. Data processor
Answer: b. Extra convolutional predictor layersd2l+1
SSD models are used in which of the following areas?
a. Autonomous driving
b. Video surveillance
c. Robotics
d. All of the above
Answer: d. All of the aboveexamtopics+1

Blog

Systemethic Technologies

Leave a Reply Cancel reply

Useful links

Other Terms & Policies

Contact Info