Re:Counting Bees on a Raspberry Pi with a ConvNet
by amelius on Friday May 18, 2018 @07:01PM
(#17105621)
Say you'd train this on images of bees of some size. Would the same net also work for bees which appear to be (e.g.) twice as large?
[ Parent
| on HN ]
Re:Counting Bees on a Raspberry Pi with a ConvNet
by mat_kelcey on Friday May 18, 2018 @11:43PM
(#17106542)
no, it won't directly. conv nets handle translation invariance but not scale invariance. having said that there's no reason you can't use aggressive data augmentation for this (resizing before patch sampling). i wonder how much the semi supervised approach might help too; if you've labelled _only_ small bees in a subset of the data, trained a model, applied to a larger dataset & retrained there will be a small amount of detections (that are true positives) to bees that are slightly larger (and smaller) than the ones you labelled.... (maybe?)
Re:Counting Bees on a Raspberry Pi with a ConvNet
by amelius on Saturday May 19, 2018 @05:56AM
(#17107257)
Ok. Why isn't there a kind of network function that works with scaling, like convolution works with translation?
Re:Counting Bees on a Raspberry Pi with a ConvNet
by alexcnwy on Saturday May 19, 2018 @08:55AM
(#17107702)
I think a CNN can handle scale invariance e.g. With max pooling
Re:Counting Bees on a Raspberry Pi with a ConvNet
by Maybestring on Friday May 18, 2018 @10:05PM
(#17106245)
Maybe, but you shouldn't count on it. Perhaps a big bee is supposed to be counted as two small bees, you didn't specify when you trained it.