Deep Learning for Computer Vision · Quiz 2 · Jan 2026

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

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

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

MobileNetV1 reduces computation primarily by using:

EfficientNet’s key scaling idea is:

You have a pretrained CNN backbone and only **500** labeled images for a new task. A strong first baseline is:

When fine-tuning a model, what is a common way to prevent the new training from "overwriting" or destroying the useful features already learned during pre-training?

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

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

Which statement is most accurate?

RetinaNet is notable primarily because it introduced:

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

Vanishing gradients in vanilla RNNs are largely caused by:

Which statement is true?

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

Which statements are true?

Which statements are true ?

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.

Using the same setup (Input feature map: 14×14×128, output: 14×14×256, kernel: 3×3, stride 1, same padding, ignore bias), compute total **MACs operation** (multiply-accumulates). A single MAC operation multiplies two numbers and adds the result to an accumulator.

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 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.

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.

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.