Convolutional Neural Networks

Computer Vision

Computer Vision Problems include:

  • Image Classication
  • Object Detection
  • … …

One of the challenges of computer vision problems is that the input can be very big. For example, a 1000 by 1000 image can have $1000 \times 64 \times 3 = 12288$ dimensions because there are three color channels. If the size of hidden layer is 1000, the number of parameters from input layer to hidden layer could be 3 billion. This will cause these problems:

  1. data size requirements;
  2. computational requirements;
  3. memory requirements.

Padding

The problems of convolutional operation:

  1. shrinking output
  2. throwing away a lot of information from the edges of the image

In order to fix these problems, what we need to do is pad the image.