Replies: 1 comment
-
Here are some of the types of cues (with examples) that we use while flying observational surveys in the Salish Sea during the winter. These aren't necessarily rules, but are hints that help the observer to quickly narrow down to potential species/taxa. Some of these are repeats from Ben's post but may help in thinking about additional processes that could be added to future models or post-model processing.
|
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
A primary use case, especially for the bird monitoring arena, is replacing existing human or image-annotation surveys. Rather than throwing away the human and trying to directly slot in a machine learning model, we want to move towards a future of collaboration. Humans know alot of things that are auxiliary to the image pixels, the behavior, location, relative abundance of species, the differences in appearance through time. All of these things can be learned by an algorithm, but in the open set problem of novel datasets, we want to reduce the burden on annotation. Fully bayesian networks are probably still immature, but I can imagine a system that builds in 1) species distributions, for example using map of life or ebird. Lots of papers from the iNaturalist community that do this well. 2) Relative abundances and co-occurrence matrices. 3) Habitat descriptions. These pieces of data could be specific to each class, or be tailored to a survey. In the best case, the human integration is minimally invasive, and operates on features or predictions of existing networks.
Beta Was this translation helpful? Give feedback.
All reactions