Neural Networks I
What is the name for the procedure used to train neural networks?
Backpropagation
Which edge detection algorithm uses two thresholds?
Canny
How many different intensity levels can be represented for one pixel in an 8-bit grayscale image?
256
Is SIFT accepted immediately after it came out?
NO
What does GAN stand for?
Generative Adversarial Network
Name the rectifying function (named for its shape) used to train early neural networks.
Sigmoid
Which edge detection algorithm uses the following filters: [-1 -2 -1; 0 0 0; 1 2 1] and [-1 0 1; -2 0 2; -1 0 1]
Sobel
How many bytes per pixel are typically used to store an RGB color image, uncompressed?
Three (one each for red, green, and blue)
What does SIFT stand for?
Scale Invariant Feature Transform
Name the two important parts of GAN
Generator & Discriminator
What was an early name for a neural network cell?
A perceptron
(Open-ended) From your own perspective, can you identify what kind of images will not give good results with common edge detectors?
e.g. Low contrast between background color and object color; too many details
What do the three components of HSV color space each represent?
Hue, saturation, and color value
SURF is an efficient algorithm that is the same as SIFT performance and reduced computational complexity.
(Open-ended) Give some examples of GAN's application
generate human faces, generate cartoon characters...
What does ReLU stand for?
Rectifying linear unit
Why the Canny method is less likely than the other methods to be fooled by noise?
This method uses two thresholds to detect strong and weak edges, including weak edges in the output if they are connected to strong edges. Thus it is more likely to detect true weak edges.
In LAB color space, which component is equivalent to a grayscale image?
The L component
What name is given to the 2D array that holds the sum of the intensities of all pixels above and to the left of a particular image location?
Integral image
How does the answer to question 2 of this section in GAN interact with each other?
The generator learns to generate plausible data. The generated instances become negative training examples for the discriminator. The discriminator learns to distinguish the generator's fake data from real data.
A neuron’s activation level is determined by the input from its connected neurons, plus what else?
The bias
Canny’s three criteria for a good edge detector are low error rate, minimal response, and what else?
Good localization
What is the supposed advantage of LAB color space over RGB?
It is supposed to be perceptually uniform
Which of the following is SIFT not designed to be invariant to: scale, occlusion, orientation, and lighting?
Occlusion
How is GAN model trained?
GAN training proceeds in alternating periods: The discriminator trains for one or more epochs. The generator trains for one or more epochs. Repeat steps 1 and 2 to continue to train the generator and discriminator networks.