CS 451 Quiz 32
Computer vision and convolution
Which of the following is NOT a computer vision problem?
Neural style transfer
If we used a 1-megapixel color image as input to a fully-connected neural net with 1000 hidden nodes in the first layer, how many parameters would the matrix W have?
The first edge detection example in the video uses a 6x6 image and a 3x3 filter. What is the size of the output image?
3 x 3
4 x 4
5 x 5
6 x 6
7 x 7
Which 3x3 filter (in Octave/Matlab notation) could we use to detect horizontal edges?
[1 0 -1; 1 0 -1; 1 0 -1]
[1 1 1; 0 0 0; -1 -1 -1]
[1 0 0; 0 1 0; 0 0 1]
Convolution "places" a filter F on each pixel of the input image. Let R denote the region of the image "covered" by the current placement. How is the resulting output pixel computed (in Octave/Matlab notation)?
sum(sum(F + R))
sum(sum(F * R))
sum(sum(F .* R))
sum(sum(F ./ R))
Why is padding useful?
So image sizes are powers of 2
So the convolution operation doesn't shrink the image
For increased efficiency
For "same" convolutions with a 5x5 filter, how many pixels of padding do we need to add on each side?
Suppose we convolve a large image with a 9x9 filter, with stride 1 or 2, and padding 0 or 3. Which option will result in the least amount of computation?
stride 1, padding 0
stride 1, padding 3
stride 2, padding 0
stride 2, padding 3
If we use "same" convolution to convolve a NxNx3 color (RGB) image with a FxFx3 filter, what are the dimensions of the output?
N x N x 1
N x N x 3
F x F x 1
F x F x 3
How clear is your understanding of convolution? (No wrong answer :)
After watching the videos, I have a pretty clear idea of what's going on
I didn't watch all the videos, so I'm confused
Even after carefully watching the videos, I'm still confused
This content is neither created nor endorsed by Google.
Terms of Service