Train a neural network to look at photographs and make a reasonable guess at the age of the subject. The training data set was collected from the IMDB website and contains over 6000 photos sorted by the age.
age-net
|-img
| |-16
| |-0a98a69aef9545d28d5743ee7f776f05.png
| |-1ad2a212a1b7419bbb363163499a7099.png
| ...
| |-efcada2b3f2a4486a6c46fa5d5ff3e15.png
| |-17
| |-18
| ...
| |-85
This project uses a fairly standard feedforward neural network based on the multilayer perceptron. It is set up to support a configurable number of layers and a configurable number of nodes in each layer.
3996 input nodes
1111 hidden nodes
999 hidden nodes
555 hidden nodes
1 output nodes
The inputs are simply the brightness values of each individual pixel in the 54 by 74 images, scaled to a value between 0.0f and 1.0f. No more image preprocessing is performed here... mostly because I don't know how to do it.
The outputs are simply the age, divided by 100, scaled between -1.5 and 1.5.