chips
building a training set, the trouble with machine learning
These caught my eye a few weeks ago when I was building a training set to define a pattern, the idea was to use a quick and dirty routine to crop 40x40 sub images. This way I could generate large amounts of data that would be mostly right and would leave me with far fewer little windows to sort through, I could then use the sets to build a much better and faster running classifier |
|
