Scientific article in PNAS: Deep learning and computer vision will transform entomology

Ecology meets AI - an interesting read about automated monitoring, based on image identification using machine learning.

Not surprisingly, iNaturalist (and its limitations) is also mentioned.

Open Access: https://www.pnas.org/content/118/2/e2002545117

4 Likes

while I’d ask about peoples’ opinions as to the ability of AI to meet or surpass humans’ ID ability, it’s been rehashed on the forum many times already…
…so, what tedious research work OTHER than identification do you see AI as capable of taking over? And do you look forward to it?

1 Like

What research work do you see AI as NOT capable of taking over? :wink:
I struggle to imagine anything that won’t be automated long term.
Roll on Neuralink I say - humans v2.0 ftw! :laughing:

Nice article. Could iNaturalist be adapted as they say?
Not sure how specific the context is at that point in the article, but I thought at least that it was explicitly against the rules to just spam the platform with camera trap content. I’m hoping to experiment with camera traps on insects a bit this year… so curious what the limitations actually are.
I guess one day, if the ML model could be used outside of iNaturalist somehow then folks could get the autosuggests without spamming the platform…

1 Like

https://www.inaturalist.org/computer_vision_demo

2 Likes

yes… I’ve come across that.
there’s no equivalent that can be used through the API or something though, to just set it to run on a batch of images from a camera trap, right?
( or any plan to make one? )

2 Likes

The next step beyond ID: selecting which characters to use in making a dichotomous key or a phylogeny. I look forward to it if it will make available keys to taxa that I currently cannot find keys for.

3 Likes

I doubt they would want to open up an API, as it would result in a lot of computationally intensive traffic that wouldn’t end up benefitting iNaturalist. It probably wouldn’t be hard to make the model downloadable and usable by people who have experience with tensorflow, but they must have a reason for holding it back.

3 Likes

It’s closed source. I suspect they licensed prewritten vision libraries rather than trying to rewrite that level of specialized code.

It is one of only a very few pieces of the ecosystem not open sourced (I don’t know the licensing of the mobile apps code).

1 Like

My understanding, with the huge caveat that I have no expertise with neural networks whatsoever, is that they would be very bad at identifying specific structural features to make a key from. Because a computer doesn’t break down an organism into structural features and classify them based on those features it does not necessarily follow that it would immediately be good at making a key that does just that. And why would you need to when the hivemind can ID it for you.:upside_down_face:

I think that’s why the deepmind images come out so weird and why I can’t always verbalize the difference between two ducks even though I know because of the way they are.

2 Likes

From the article:

One of the main challenges with deep learning is the need for large amounts of training data, which is slow, difficult, and expensive to collect and label.

Indeed, and maybe too much for the process to be feasible. Taxonomists are in short, and declining, supply. Many large genera have never been subjected to a modern revision, and Museum drawers are full of undescribed species. Take Agra for example: https://www.inaturalist.org/taxa/251408-Agra.

It is not matter of machines “collecting” training data, but of human experts producing them manually - well, intellectually.

1 Like

From what I’ve read about Visipedia, which is what iNat’s model is based on, (at least with birds) one step of the model is segmenting the body parts of the organism.

https://biodiversityanalysis.nl/files/User%20manual%20Nature%20Identification%20API%20-%20Public.pdf
yes tgere was
naturalist and a post 1month ago.
Api was already gone
I now provicnes who do wild camera traps

https://www.naturetoday.com/intl/nl/nature-reports/message/?msg=26867

https://www.utwente.nl/en/designlab/news/2020/6/650650/spotting-wild-boars-by-using-drones

https://www.rtvoost.nl/nieuws/1322551/Slimme-camera-s-gaan-provincie-helpen-bij-het-tellen-van-wild

wildspotter.nl
https://www.rtvoost.nl/nieuws/250462/Overijssel-helpt-massaal-mee-met-wild-spotten

https://biodiversityanalysis.nl/files/User%20manual%20Nature%20Identification%20API%20-%20Public.pdf

https://www.wildspotter.nl/ Is already gone, vanished
ttps://identify .biodiversityanalysis .nl /beta/observation/demo. Is already gone, vanished

1 Like

This was the old one https://identify.biodiversityanalysis.nl/beta/observation/demo

1 Like

I’ll do it. If I have assistance in paying my bills, that is. :grin: