I was asked in the comments to last week’s blog post if I could provide some feedback about the results of Season 4. If you felt like you were seeing a lot of “nothing here” images, you’re right: of the 158,098 unique capture events we showed you, 70% were classified as having no animals in them. That left 47,320 with animals in them to classify, and the vast majority of these (94%) contained just one species. Here’s the breakdown of what was in all those images:
Maybe it won’t surprise you that Season 4 covered 2012’s wet season, when over a million wildebeest, zebra, and Thomson’s gazelle migrate through our study area. I find it interesting that hartebeest are also pretty numerous, but I wonder if it’s because of that one hartebeest that stood in front of the camera for hours on end.
This pie chart is based on the number of what we call “capture events,” which is the set of 1 or 3 pictures you see every time you make a classification. Once a camera has taken a set of pictures, we delay it from triggering again for about a minute. That way we don’t fill up the camera’s memory card with too many repeats of the same animals before we have a chance to replace them. But a minute isn’t a very long time for an animal that has decided to camp out in front of a camera, and so we frequently get sequences of many capture events that are all of the same animal. One of the things we’ll have to do in turning your classifications into valid research results is to figure out how to find these sequences in the data automatically.
Here’s a sequence of an elephant family hanging out around our camera for the night about a year ago. (Hat tip to dms246 who put together a collection of most of these images to answer the concerned question of some classifiers who saw just one image out of the whole sequence: is that elephant dead or just sleeping?)
If you’re interested in how I made the above pie chart, keep reading. But we’re going to get technical here, so if algorithms don’t interest you, feel free to stop.