Back to Subject

Unit 3: Transfer Learning and Fine-Tuning for Vision Tasks - Practice Quiz

CSE471 — Deep Learning For Computer Vision 60 Questions
0 Correct 0 Wrong 60 Left
0/60

1 What is transfer learning in computer vision?

Transfer learning Easy
A. Training a model without any data
B. Converting images into text files
C. Removing all layers from a neural network
D. Using knowledge from one task for another task

2 Why is transfer learning commonly used for image classification?

Transfer learning Easy
A. It requires fewer labeled training images
B. It removes the need for an output layer
C. It guarantees perfect classification accuracy
D. It prevents images from having different sizes

3 What is the main feature of ResNet architectures?

ResNet Easy
A. A complete absence of pooling layers
B. Only one convolutional layer
C. Training without labeled examples
D. Shortcut connections between layers

4 What problem do residual connections in ResNet help address?

ResNet Easy
A. Difficulty training very deep networks
B. Missing class names in datasets
C. Excessive image brightness
D. Lack of image file formats

5 What is a key idea behind EfficientNet?

EfficientNet Easy
A. Using only fully connected layers
B. Training every model with one fixed image size
C. Replacing images with numerical labels
D. Scaling depth, width, and resolution together

6 EfficientNet models are designed mainly to improve which balance?

EfficientNet Easy
A. Labels and file compression
B. Accuracy and computational efficiency
C. Batch size and class names
D. Color and image brightness

7 What is MobileNet primarily designed for?

MobileNet Easy
A. Processing only text-based documents
B. Training models with unlimited memory
C. Efficient use on mobile and embedded devices
D. Replacing all image preprocessing steps

8 Which operation is commonly used by MobileNet to reduce computation?

MobileNet Easy
A. Depthwise separable convolution
B. Three-dimensional pooling only
C. Unrestricted fully connected layers
D. Repeated image enlargement

9 In feature extraction, what is usually done with the pretrained convolutional layers?

Feature extraction Easy
A. Their layers are randomly removed
B. Their weights are deleted
C. Their inputs are replaced with labels
D. Their weights are kept fixed

10 What do early convolutional layers often learn from images?

Feature extraction Easy
A. Database connection rules
B. Final class names only
C. Edges and simple textures
D. Complete sentences

11 When using a pretrained model as a feature extractor, which part is commonly replaced?

Feature extraction Easy
A. The final classification layer
B. Every convolutional filter
C. The image file extension
D. The training dataset format

12 What does fine-tuning mean in transfer learning?

Fine-tuning Easy
A. Updating some pretrained weights on new data
B. Freezing the entire neural network permanently
C. Using a model without any training
D. Changing image labels without changing model weights

13 Fine-tuning is often useful when the target dataset is:

Fine-tuning Easy
A. Related but different from the pretraining dataset
B. Completely empty of images
C. Made only of numerical formulas
D. Identical to the model architecture

14 What does it mean to freeze a neural network layer?

Managing freezing and unfreezing of layers Easy
A. Its filters are converted into labels
B. Its weights are not updated during training
C. Its input images are permanently resized
D. Its outputs are always set to zero

15 Why are pretrained layers often frozen at the beginning of training?

Managing freezing and unfreezing of layers Easy
A. To remove the need for a loss function
B. To make every layer randomly initialized
C. To increase the number of target classes
D. To preserve useful learned features

16 During gradual fine-tuning, what is commonly done after training the new classifier?

Managing freezing and unfreezing of layers Easy
A. The model output is changed to audio
B. Some earlier layers are unfrozen
C. All input images are discarded
D. The loss function is removed

17 What is a layer-wise learning rate strategy?

Applying layer-wise learning rates Easy
A. Assigning one label to every layer
B. Using different learning rates for different layers
C. Using one batch for every layer
D. Changing the image size after every layer

18 Which learning rate is commonly used for newly added classification layers?

Applying layer-wise learning rates Easy
A. A smaller rate than every frozen layer
B. A rate that never changes model weights
C. A permanently zero learning rate
D. A relatively larger learning rate

19 What is domain adaptation used for?

Domain adaptation and dataset Easy
A. Removing all differences between image classes
B. Converting a classification task into file storage
C. Adapting a model to a different data distribution
D. Increasing the number of network parameters automatically

20 What is the target dataset in transfer learning?

Domain adaptation and dataset Easy
A. A dataset that has no relationship to the task
B. A dataset containing model source code
C. The dataset used for the new task
D. The dataset used only to initialize random numbers

21 A team has 2,000 labeled images of industrial defects and a model pretrained on ImageNet. What is the most appropriate initial transfer-learning strategy?

Transfer learning Medium
A. Replace all convolutional layers with fully connected layers
B. Freeze the classifier and retrain only the earliest convolution
C. Train the entire network from randomly initialized weights
D. Freeze the backbone and train a new task-specific classifier

22 Transfer learning from ImageNet is likely to provide the smallest benefit for which target task?

Transfer learning Medium
A. Classifying patterns in radar measurements that differ substantially from natural images
B. Identifying species in wildlife photographs
C. Classifying common household objects
D. Recognizing vehicles in street images

23 A 50-layer plain CNN trains poorly, while a ResNet-50 trains successfully on the same dataset. Which ResNet mechanism most directly explains the improvement?

ResNet Medium
A. Residual connections provide shorter paths for gradient flow
B. Max pooling prevents every vanishing-gradient problem
C. Dropout is automatically added after each convolution
D. Residual blocks remove the need for nonlinear activations

24 In a ResNet block, the residual branch produces and the shortcut carries . When their shapes match, what is the block output before the final activation?

ResNet Medium
A.
B.
C.
D.

25 A practitioner wants to scale an EfficientNet model while preserving its core scaling principle. Which change best follows compound scaling?

EfficientNet Medium
A. Increase width and remove all resolution scaling to keep the model's spatial dimensions unchanged
B. Increase only depth while keeping width and resolution fixed
C. Increase depth, width, and input resolution in a balanced way
D. Increase only input resolution until GPU memory is exhausted

26 When replacing EfficientNet-B0 with a larger EfficientNet variant for transfer learning, what trade-off should be expected?

EfficientNet Medium
A. Identical accuracy with lower computation and memory use
B. Fewer parameters because every larger variant uses stronger pooling
C. Potentially better accuracy with greater computation and memory use
D. No change in inference latency because compound scaling affects training only

27 Why is MobileNet commonly selected for an image-classification application running on a mobile device?

MobileNet Medium
A. Every convolution operates only on one-pixel input images
B. Depthwise separable convolutions reduce computation and parameters
C. Large fully connected layers improve low-power inference
D. Residual additions eliminate the need to store feature maps

28 A MobileNet model uses a width multiplier . What is the primary effect of decreasing ?

MobileNet Medium
A. It reduces channel counts and computational cost
B. It freezes a larger fraction of pretrained layers
C. It converts depthwise convolutions into standard convolutions with additional trainable kernels
D. It increases image resolution and receptive-field size

29 A pretrained CNN is used only to compute embeddings, which are then supplied to a newly trained linear classifier. Which layer output is usually the most suitable embedding?

Feature extraction Medium
A. The raw RGB image before normalization
B. The loss value computed after each training batch
C. The original classifier's class-probability vector
D. The final pooled feature vector before the original classifier

30 A frozen pretrained backbone is used for feature extraction, but the new classifier overfits. Which intervention most directly addresses the problem without modifying backbone weights?

Feature extraction Medium
A. Add regularization and augmentation while training the classifier
B. Unfreeze every backbone layer and greatly increase its learning rate
C. Remove the validation set to provide more classifier training samples
D. Cache each training feature multiple times and treat the copies as independently labeled examples

31 After training a new classification head, a practitioner unfreezes part of the pretrained backbone. Which learning-rate change is generally appropriate?

Fine-tuning Medium
A. Use the largest stable learning rate for all layers
B. Set the backbone learning rate permanently to zero
C. Use a smaller learning rate for the unfrozen backbone
D. Increase the learning rate whenever validation loss rises

32 During fine-tuning, training accuracy rises while validation accuracy falls after several epochs. What is the most appropriate response?

Fine-tuning Medium
A. Increase model size without changing the training setup
B. Continue training until both accuracies reach 100%
C. Apply early stopping and strengthen regularization
D. Evaluate only on the training set because pretrained models do not require validation

33 For a small target dataset similar to ImageNet, which staged training procedure is generally most appropriate?

Managing freezing and unfreezing of layers Medium
A. Train the new head, then unfreeze later blocks gradually
B. Unfreeze the entire network before training the new head
C. Train only the earliest convolution and discard later blocks
D. Alternate frozen and unfrozen states after every individual batch throughout training

34 Why are the final backbone blocks usually unfrozen before the earliest blocks?

Managing freezing and unfreezing of layers Medium
A. Earlier blocks contain no trainable parameters
B. Later blocks always require less GPU memory
C. Later blocks contain more task-specific semantic features
D. Earlier blocks produce class probabilities rather than visual features

35 A frozen backbone contains Batch Normalization layers. The validation performance becomes unstable when these layers continue updating their running statistics on very small batches. What should be tried?

Managing freezing and unfreezing of layers Medium
A. Increase the classifier learning rate after every batch
B. Replace running means with randomly sampled values during each validation pass
C. Keep the Batch Normalization layers in evaluation mode
D. Delete all normalization layers from the pretrained network

36 A model is divided into an early backbone, a late backbone, and a new classifier. Which learning-rate assignment best matches discriminative fine-tuning?

Applying layer-wise learning rates Medium
A. , , and , respectively
B. , , and , respectively
C. , , and , respectively
D. , , and , respectively

37 Why can layer-wise learning rates reduce catastrophic forgetting during transfer learning?

Applying layer-wise learning rates Medium
A. They limit changes to general features in early layers
B. They guarantee that the target dataset has balanced classes
C. They prevent the classifier from receiving any gradients
D. They force all layers to converge to identical parameter values after each optimization step

38 A classifier is trained on studio product photos but deployed on dim warehouse-camera images. Which change most directly reduces this domain shift?

Domain adaptation and dataset Medium
A. Shuffle the studio training set using a different random seed
B. Increase the number of source-domain output classes
C. Add more studio photos with the same lighting
D. Fine-tune using labeled warehouse-camera images

39 Only unlabeled target-domain images are available. Which technique can still help align source and target representations?

Domain adaptation and dataset Medium
A. Assigning every target image to the most frequent source class and treating those assignments as verified ground truth
B. Supervised fine-tuning using invented target labels
C. Training only on source labels without using target images
D. Adversarial domain adaptation with a domain discriminator

40 A medical-image dataset contains multiple scans from each patient. Which split best avoids leakage when evaluating a transfer-learned model?

Domain adaptation and dataset Medium
A. Use the same patient scans for training and testing but apply different image normalization methods
B. Randomly split individual scans across all dataset splits
C. Assign all scans from a patient to only one split
D. Place augmented scans in validation and originals in training

41 A CNN pretrained on ImageNet is transferred to a medical-image dataset with only 800 labeled examples. The target images are grayscale, but their low-level edge statistics resemble natural images. Which initialization strategy is most defensible?

Transfer learning Hard
A. Replace every convolutional filter with random weights before classifier training
B. Replicate the grayscale channel three times and retain pretrained convolutional weights
C. Convert filters to grayscale and train all layers with the same large learning rate
D. Initialize the first layer randomly and retain only the pretrained classifier weights

42 In a pretrained ResNet block, the identity shortcut has dimension while the residual branch outputs channels and uses stride 2. Which modification preserves the residual-addition operation?

ResNet Hard
A. Remove the shortcut and double the residual branch activation width
B. Add zero padding to the identity tensor without changing its spatial size
C. Use a projection with stride 2 on the identity branch
D. Average the residual channels into channels before addition

43 EfficientNet scales network depth, width, and input resolution using a compound coefficient. If compute is approximately proportional to , which scaling choice best preserves balanced growth when increasing model capacity?

EfficientNet Hard
A. Increase only width because channel expansion dominates representational capacity
B. Increase depth, width, and resolution by unrelated manually selected factors
C. Increase resolution first and keep depth and width fixed for stability
D. Increase all three dimensions under a shared compound scaling constraint

44 For an input feature map of size and a depthwise-separable convolution producing output channels with kernel size , what is the approximate parameter reduction relative to a standard convolution?

MobileNet Hard
A. From to
B. From to
C. From to
D. From to

45 A frozen backbone produces features that separate training classes well, but target validation accuracy remains poor after replacing the classifier. Which observation most strongly suggests that the fixed representation is unsuitable for the target task?

Feature extraction Hard
A. The backbone contains more parameters than the newly initialized classifier
B. The target classifier uses fewer output units than the source classifier
C. The classifier loss decreases while validation errors remain structured by target subdomain
D. The optimizer reports a lower learning rate for the classifier than for the backbone

46 A pretrained model reaches high training accuracy immediately after unfreezing all layers, but target validation accuracy falls and feature norms grow rapidly. Which intervention is most appropriate first?

Fine-tuning Hard
A. Freeze only the classifier and increase augmentation strength afterward
B. Increase the learning rate to escape the pretrained parameter basin
C. Use a smaller backbone learning rate with stronger regularization
D. Reinitialize the backbone and train it from scratch on the target set

47 During staged fine-tuning, the backbone is frozen but contains BatchNorm layers. The target dataset has a different intensity distribution and is small. Which configuration is generally safest?

Managing freezing and unfreezing of layers Hard
A. Replace BatchNorm with dropout and unfreeze every convolutional layer
B. Remove all BatchNorm layers before transferring the pretrained backbone
C. Keep BatchNorm statistics frozen while training only the new classifier
D. Update BatchNorm statistics on every target batch regardless of batch size

48 A three-stage backbone uses learning rates , , and from the earliest to latest stage, while the new classifier uses . What principle does this schedule encode?

Applying layer-wise learning rates Hard
A. Earlier features receive larger updates because they have higher generality
B. The classifier is protected from overfitting by using the largest learning rate
C. Later features and the classifier adapt more rapidly to target semantics
D. Every stage receives the same effective update after normalization

49 A source model classifies daytime street images, while the target dataset contains nighttime images with identical labels. Unlabeled target data are abundant. Which method directly addresses the domain shift without requiring target labels?

Domain adaptation and dataset Hard
A. Discard nighttime images because their labels cannot be verified
B. Align source and target feature distributions with an unsupervised adaptation loss
C. Use random labels for target images to balance the training batches
D. Train a larger classifier using only the labeled source images

50 A target dataset has the same visual domain as the pretraining data but a highly different label taxonomy. Which transfer strategy best reduces harmful source-task bias?

Transfer learning Hard
A. Reuse the entire source classifier and freeze the complete network
B. Reuse the backbone, replace the classifier, and tune upper layers cautiously
C. Train only the source classifier until its accuracy stops improving
D. Copy source logits directly into the target labels without recalibration

51 Why can a ResNet often be fine-tuned more deeply than a plain network of comparable depth, even when the target dataset is small?

ResNet Hard
A. Residual branches eliminate the need for normalization during adaptation
B. Shortcut connections guarantee that target-domain features are linearly separable
C. Residual shortcuts make every layer independent of the loss function
D. Identity paths improve gradient propagation and preserve pretrained representations

52 When transferring an EfficientNet model to a target dataset whose objects occupy much smaller image regions than in pretraining, which change is most likely to help before substantially increasing model width?

EfficientNet Hard
A. Replace depthwise convolutions with fully connected layers at every stage
B. Remove squeeze-and-excitation modules to prevent channel recalibration
C. Increase input resolution while preserving the architecture's scaling assumptions
D. Reduce input resolution so the object occupies fewer feature-map cells

53 A MobileNet-based classifier must run under a strict latency budget. Validation accuracy improves after increasing width, but measured latency exceeds the budget. Which adjustment is most principled?

MobileNet Hard
A. Reduce the width multiplier and evaluate accuracy at the deployment resolution
B. Increase width further and compensate with heavier quantization later
C. Replace depthwise convolutions with standard convolutions for better hardware use
D. Add dense layers because parameter count alone determines inference latency

54 A frozen backbone is used with global average pooling. The target classes differ mainly by precise spatial arrangement rather than overall texture. What limitation should be expected?

Feature extraction Hard
A. Global pooling makes the feature dimension dependent on image width
B. Global pooling always increases sensitivity to object translation errors
C. Global pooling prevents gradients from reaching the classifier parameters
D. Global pooling may discard spatial information needed to distinguish arrangements

55 A target dataset contains severe class imbalance. During fine-tuning, overall accuracy rises while minority-class recall collapses. Which evaluation and optimization change is most appropriate?

Fine-tuning Hard
A. Reduce augmentation because minority examples should remain visually identical
B. Use balanced metrics and class-aware loss or sampling during adaptation
C. Use accuracy alone because it reflects the deployment class frequencies
D. Freeze the classifier and optimize only the backbone with unweighted loss

56 After unfreezing a new block, its gradients are nonzero but its weights appear unchanged during training. Which implementation issue is most likely?

Managing freezing and unfreezing of layers Hard
A. The classifier has a larger output dimension than the backbone
B. The input images were normalized using source statistics
C. The residual block contains an identity shortcut
D. The block is absent from the optimizer parameter groups

57 Two layers have gradients with similar magnitudes, but the earlier layer has much larger pretrained weights and is more transferable. Why can assigning it a smaller learning rate still be justified?

Applying layer-wise learning rates Hard
A. Layer-wise rates remove the need to monitor validation performance
B. Learning rates determine only gradient direction, not parameter displacement
C. Smaller rates limit relative parameter drift and protect established features
D. Smaller rates guarantee identical updates across layers with different norms

58 A target validation set was collected from the same camera and locations as the training set, while deployment uses unseen cameras. Fine-tuning improves validation accuracy but deployment performance decreases. What is the primary diagnosis?

Domain adaptation and dataset Hard
A. The source dataset must be discarded because deployment has unseen cameras
B. The target classifier requires more output classes than the deployment system
C. The backbone has become too shallow to represent camera-specific features
D. The model has overfit the validation distribution and the evaluation is not representative

59 A target dataset is larger than the source pretraining dataset but differs substantially in sensor modality and image formation. Which conclusion is most defensible?

Transfer learning Hard
A. The target dataset size alone guarantees that transfer will improve accuracy
B. The pretrained model should always be frozen because modality changes are irreversible
C. Source pretraining is useless whenever the target dataset has more examples
D. Transfer may still help, but modality mismatch should be tested against strong baselines

60 A fine-tuned model performs well on clean target images but fails under a mild illumination change that is common in deployment. Which response best targets the failure mode?

Fine-tuning Hard
A. Remove validation images with illumination changes to stabilize the metric
B. Increase classifier depth without changing the training distribution
C. Use augmentation or domain-relevant normalization that represents illumination variation
D. Freeze all layers permanently so illumination cannot affect learned features
Previous MCQ
Unit 2
Next MCQ
Unit 4