Computer Vision

Example of computer vision problems:

Challenges

• Inputs can be really big.

• This example, gives $64*64*3=12288$

• This example, gives $1000*1000*3=3,000,000$
  1,000x1,000 pixels

So, in the NN, input ($x\in \mathbb{R}^{3M}$) will be 3 million dimention (3000000, 1) and assuming the first hidden layer has 1000 units, then $W^{[1]}$ dimention is (1000, 3000000) and this matrix will have 3 billion parameters.