Revision history [back]

size is basically what it sounds like. How many pixels around the center do you use when calculating the keypoint? In some methods, this is variable.

angle is if you calculate an orientation. If you have a way of telling which way is "forward", for some consistent definition. This is often not used.

response is how strong your detector's belief that this is a keypoint is. So you can sort by strength. You should have something like this, even if it's just the FAST or corner response from earlier in your method.

octave is if you have multiple scales, which one is it found on? If you build a pyramid of images and search for keypoints on all of them, which one is this point from?

class_id: I don't know. I've never seen anything use this. I suppose if you have more than one distinct type you can put that here?