Deep Learning for Computer Vision · Quiz 2 · May 2026

← Course papers · Start practice / exam

Questions and published explanations below are available without starting a test. Some questions may not have a published solution yet.

0 / 22 published solutions

Question 2 MCQ · 1.0 marks

Why do very deep plain CNNs (without skip connections) sometimes show **higher training error** than shallower CNNs?

6406537126022.

They always overfit because parameter count increases.

6406537126023.

Optimization becomes harder; skip connections make learning identity mappings easier and improve gradient flow.

6406537126024.

BatchNorm guarantees this never happens.

6406537126025.

Using sigmoid activations instead of ReLU always fixes it.

A published solution is not available for this question yet.

Report a correction for Q2

Question 3 MCQ · 1.0 marks

MobileNetV1 reduces computation primarily by using:

6406537126034.

Dilated convolutions everywhere

6406537126035.

Depthwise separable convolutions (depthwise + pointwise)

6406537126036.

ROI pooling

6406537126037.

Fully-connected layers instead of convolutions

A published solution is not available for this question yet.

Report a correction for Q3

Question 4 MCQ · 1.0 marks

EfficientNet’s key scaling idea is:

6406537126038.

Only scale depth while keeping width and resolution fixed

6406537126039.

Only scale resolution while keeping depth and width fixed

6406537126040.

Compound scaling of depth, width, and resolution using a single coefficient

6406537126041.

Scale width but reduce depth to keep parameters constant

A published solution is not available for this question yet.

Report a correction for Q4

Question 5 MCQ · 1.0 marks

Which technique most directly helps reduce catastrophic forgetting early in fine-tuning?

6406537126042.

Use very high weight decay on the backbone

6406537126043.

Use differential learning rates (small LR for backbone, larger LR for head)

6406537126044.

Replace ReLU with sigmoid

6406537126045.

Remove skip connections

A published solution is not available for this question yet.

Report a correction for Q5

Question 6 MCQ · 1.0 marks

Which method produces a heatmap by weighting convolutional feature maps using gradients of a target class score?

6406537126046.

PCA

6406537126047.

Grad-CAM

6406537126048.

Histogram equalization

6406537126049.

t-SNE

A published solution is not available for this question yet.

Report a correction for Q6

Question 7 MCQ · 1.0 marks

Which detector is a classic **two-stage** detector with an RPN + ROI feature extraction?

6406537126050.

SSD

6406537126051.

YOLO

6406537126052.

Faster R-CNN

6406537126053.

RetinaNet

A published solution is not available for this question yet.

Report a correction for Q7

Question 8 MCQ · 1.0 marks

RetinaNet is notable primarily because it introduced:

6406537126058.

ROI Align

6406537126059.

Focal loss to address class imbalance

6406537126060.

Depthwise separable convolutions

6406537126061.

Self-attention

A published solution is not available for this question yet.

Report a correction for Q8

Question 9 MCQ · 1.0 marks

A model that predicts a class label for **every pixel** without separating object instances is:

6406537126062.

Object detection

6406537126063.

Instance segmentation

6406537126064.

Semantic segmentation

6406537126065.

Captioning

A published solution is not available for this question yet.

Report a correction for Q9

Question 10 MCQ · 1.0 marks

Which statement is true?

6406537126070.

Transformers rely on recurrence to model sequences.

6406537126071.

Transformers cannot handle variable-length sequences.

6406537126072.

Self-attention enables direct interactions between any pair of tokens within a layer.

6406537126073.

Transformers require optical flow for vision tasks.

A published solution is not available for this question yet.

Report a correction for Q10

Question 11 MCQ · 2.0 marks

In a standard ResNet-50 bottleneck block, the three convolutions are typically:

6406537126026.

3×3, 3×3, 3×3

6406537126027.

1×1, 3×3, 1×1

6406537126028.

5×5, 3×3, 1×1

6406537126029.

1×1, 1×1, 3×3

A published solution is not available for this question yet.

Report a correction for Q11

Question 12 MCQ · 2.0 marks

In Inception/GoogLeNet modules, the main purpose of 1×1 convolutions is to:

6406537126030.

Increase spatial resolution

6406537126031.

Reduce channel dimensionality (bottleneck) and add non-linearity

6406537126032.

Replace pooling

6406537126033.

Implement residual learning

A published solution is not available for this question yet.

Report a correction for Q12

Question 13 MCQ · 2.0 marks

Which statement is most accurate?

6406537126054.

Two-stage detectors do not use CNN backbones.

6406537126055.

Single-stage detectors cannot regress bounding boxes.

6406537126056.

Two-stage detectors typically generate proposals then classify/refine

6406537126057.

Single-stage detectors require ROI pooling.

A published solution is not available for this question yet.

Report a correction for Q13

Question 14 MCQ · 2.0 marks

Vanishing gradients in vanilla RNNs are largely caused by:

6406537126066.

Too much data augmentation

6406537126067.

Repeated multiplication by Jacobians whose spectral norm is often < 1

6406537126068.

Using convolutions before the RNN

6406537126069.

Using attention layers

A published solution is not available for this question yet.

Report a correction for Q14

Question 15 MSQ · 3.0 marks

Which statements can be true in practice when fine-tuning with small batch sizes?

6406537126074.

Freezing BatchNorm running statistics can improve stability.

6406537126075.

If the new dataset distribution differs, re-estimating BN stats may help.

6406537126076.

BN always improves performance even with batch size 1.

6406537126077.

Setting BN to eval disables learned affine parameters (γ, β).

A published solution is not available for this question yet.

Report a correction for Q15

Question 16 MSQ · 3.0 marks

Which statements are true ?

6406537126078.

Smooth L1 (Huber) is often used for box regression due to robustness to outliers.

6406537126079.

Focal loss primarily addresses foreground/background (class) imbalance.

6406537126080.

IoU loss cannot be used for box regression.

6406537126081.

In two-stage detectors, the RPN has its own objectness + box regression losses.

A published solution is not available for this question yet.

Report a correction for Q16

Question 17 MSQ · 3.0 marks

Which statements are true?

6406537126082.

Soft attention is differentiable and can be trained with backpropagation.

6406537126083.

Hard attention often needs sampling + REINFORCE (or similar) due to non- differentiability.

6406537126084.

Hard attention always has higher test-time compute than soft attention.

6406537126085.

Cross-attention can align decoder queries to encoder keys/values (e.g., words attending to image regions).

A published solution is not available for this question yet.

Report a correction for Q17

Question 18 NAT · 4.0 marks

Input feature map: 14×14×128, output: 14×14×256, kernel: 3×3, stride 1, same padding, ignore bias. Compute **number of parameters** for a standard convolution layer.

A published solution is not available for this question yet.

Report a correction for Q18

Question 19 NAT · 4.0 marks

Input feature map: 14×14×128, output: 14×14×256, kernel: 3×3, stride 1, same padding, ignore bias, compute total **MACs** (multiply-accumulates).

A published solution is not available for this question yet.

Report a correction for Q19

Question 20 NAT · 4.0 marks

A ResNet bottleneck block takes 256 channels in and uses: 1×1 conv to 256 channels → 3×3 conv at 256 → 1×1 conv to 1024 channels. Ignore bias. Compute total parameters in these **three convolutions** (exclude projection shortcut).

A published solution is not available for this question yet.

Report a correction for Q20

Question 21 NAT · 4.0 marks

A detector uses three pyramid levels: P3 = 80×80, P4 = 40×40, P5 = 20×20. At each location, it uses 9 anchors (3 scales × 3 aspect ratios). Compute total anchors per image.

A published solution is not available for this question yet.

Report a correction for Q21

Question 22 NAT · 4.0 marks

Box A: top-left (2,2), bottom-right (10,12) Box B: top-left (6,5), bottom-right (14,15) Compute IoU rounded to 3 decimals.

A published solution is not available for this question yet.

Report a correction for Q22

Question 23 NAT · 4.0 marks

Let query q = [1,2]^T and keys:k1 = [1,0]^T, k2 = [0,1]^T, k3 = [1,1]^T. Scores si = q^T ki. Use softmax over scores. Use exp(1)=2.72, exp(2)=7.39, exp(3)=20.09. What is α3 (the attention weight for k3)? Provide a value in a reasonable range.

A published solution is not available for this question yet.

Report a correction for Q23