Computer Vision

Question 1

What are some uses for computer vision?

Answer 1

1. 3D Image Reconstruction
2. Object Detection
3. Image Segmentation

Question 2

How are images represented?

Answer 2

Images are represented by 3 matrices, with each of the matrices representing either Red, Green, or Blue.

Question 3

What is a basic approach to computer vision?

Answer 3

The image is represented as a vector and each pixel has it’s own input node resulting in a high number of weights for an image.

Question 4

What is a disadvantage of the basic approach?

Answer 4

1. It doesn’t scale very well as images increase in size.
2. It is not robust to small changes in the input.
3. It doesn’t take advantage of the correlations between pixels.

Question 5

What was the main development of Computer Vision after 1980?

Answer 5

The introduction of Deep Convolutional Neural Networks comes with more layers, new architectures, and new algorithms.

Question 6

What are the limitations of Deep Leanring?

Answer 6

1. DL is Data Hungry and needs a lot of data
2. DL is heavy and often needs GPUs and cloud computing to train it.
3. DL is bad at representing uncertainty and its easy to trick a NN into thinking its right.
4. It’s hard to optimise
5. Its hard to interpret as NN’s are black boxes