HN user
"The system looks at sets of pictures of the hive door taken every 10 seconds. It then extrapolates out the background, assesses the objects that have moved in the frame, and then counts the things that are likely to be bees."
"The Techcrunch article spamtastically plagiarises a BoingBoing article from the day before (including misspelling the author's name) and complete and utterly misunderstands the core technique"
Just a randomly sampled 32x32 patch, intution being most bees in image aren't bigger
That's just a sanity check; the end goal is some accelerated hardware ( neural ompute stick / jevois / etc)
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?)
yeah, that's true it's only counting bees in one image, which i thought was part 1 of any of these other things. can you post your approach / code to doing any of these things you mention? i'd love to see what you've done so far.
How is the $150 broken down?
+1
I think the Lithuania bit is a false alarm. The pipeline needs a lot more work and investigation of results. The main thing I was focusing on first was scaling it out.
oh your irony...
I was using the 2011-07-22 set so the data's already a bit stale...