Computer Vision 2

Question 1

What is imagenette?

Answer 1

Subset of ImageNet dataset created by fast.ai for faster prototyping, contains 10 classes that are easily distinguishable for humans

Question 2

How pooling works?

Answer 2

Instead of applying a kernel with learnable weights the average or maximum of the values at the kernel position is taken

Question 3

2 types of pooling

Answer 3

Either specify the kernel size (classical pooling) or the desired output size (adaptive pooling)

Question 4

What is ResNet block?

Answer 4

key building block used in ResNet (Residual Network), a deep neural network architecture that was introduced to solve the problem of vanishing gradients and to enable the training of very deep networks

Question 5

What pooling does to feature map?

Answer 5

reducing it

Question 6

Key features of ResNet block?

Answer 6

Skip connections, predicts the difference (residual) between the optimal mapping and the input

Question 7

What does ResNet to loss landscape?

Answer 7

Smoothing it: reduces the likelihood of sharp peaks and valleys, allows model to directly propagade the input forward and the gradient backwards, reducing the number of nonlinear transformations

Question 8

What is stem of the network?

Answer 8

Higher layers have more parameters (more channels) but lower layers perform more computations (greater height and width of the feature map)

Question 9

What bottleneck layers do?

Answer 9

reducing the number of channels, then, applying the
computationally more expensive convolution, and finally, increasing the number of channels to the
original size again

Question 10

What is momentum?

Answer 10

optimization technique used to improve gradient descent by adding a fraction of the previous update to the current one

Question 11

Why use momentum?

Answer 11

helps to accelerate the convergence process and smooth out updates, leading to more efficient and stable training

Question 12

What is is RMSProp?

Answer 12

optimizer that adapts the learning rate per weight

Question 13

What is Adam?

Answer 13

Combines ideas of Momentum and RSMProp in an algorithm called adaptive moment estimation

Question 14

If the number of input channels does not equal the desired number of output channels a true identity path is never possible

Answer 14

True

Question 15

A convolution with a kernel size of 1 x 1 would not make sense in the stem of a state of the are ResNet, as these convolutions do not reduce the size of the feature map.

Answer 15

True

Question 16

Bottleneck layers do not necessarily have fewer kernels than plain ResNet layers

Answer 16

True

Question 17

The reduction in number of operations from a 9 x 9 kernel to a 3 x 3 kernel is proportionally the same as from a 3 x 3 kernel to a 1 x 1 kernel if one disregards the bias related computations

Answer 17

True

Question 18

When preprocessing the dataset, the image size for batch_tfms must be less than or equal to the image size for item_tfms.

Answer 18

True