Humans algorithm possibility
-
by
clm422
For most of the humans I identify, they are adjusting the camera and as a consequence hitting or rubbing the microphone. I wonder if the screening algorithm could take this microphone sound into account to screen out the humans.
Posted
-
by
AnLand
moderator
From what I did understand, automatic filtering of sound effects is even harder than visual cues. It is easiest with highly stereotypic and repetitive sequences. I mean, I can often hear from the sound pattern that somebody is working on the camera, but for machine learning algorithms I think it is hard to find a reliable pattern in the different situations.
Maybe @akalan can explain this better and can cite a paper or similar ...
Posted
-
by
ksigler
moderator
Interesting idea, though I think it might screen out the more extreme camera reaction clips, like honey badgers mauling the camera and so forth. And there are many clips where humans are seen, but they aren't touching the camera. I don't know the proportion, but I suspect it might be of limited value/ROI compared to having us just mark them.
Posted
-
by
akalan
scientist
Hi all, AnLand is correct, human-induced sounds in these clips would be even harder to capture via automatic methods than the animal sounds because they are even more variable. Regardless, automatic visual and audio methods are not good enough to do the work Chimp&See scientists like you are doing right now, hence the reason for citizen science projects in general 😃
Posted