We passed 90,000 species in the Computer Vision Model (v2.14)!
We released a new computer vision model today. It has 1,773 new taxa (90,290 taxa up from 88,517). This new model (v2.14) was trained on data exported on May 12, 2024.
Here's a graph of the models release schedule since early 2022 (segments extend from data export date to model release date) and how the number of species included in each model has increased over time.
The graph below shows model accuracy estimates using 1,000 random Research Grade observations in each group not seen during training time. The paired bars below compare average accuracy of model 2.13 with the new model 2.14. Each bar shows the accuracy from Computer Vision alone (dark green) and Computer Vision + Geo (green). Overall the average accuracy of 2.14 is 89.2% (statistically the same as 2.13 at 89.1% - as described here we probably expect ~2% variance all other things being equal among experiments).
Here is a sample of new species added to v2.14:
- Aves
- Plantae
- Fungi
- Reptilia
- Amphibia
- Other Animalia
- Mollusca
- Mammalia
- Actinopterygii
- Insecta
- Protozoa
- Chromista
- Arachnida
We apologize for the delay in releasing v2.14. But this means that v2.15 (which we kicked off today) will probably add more than 3k species. If we can continue at this rate, we're on track to break 100,000 species in the model in early 2025!
Comentarios
I spent a fair amount of time recently to straighten out the iNat determinations of Collinsia concolor and C heterophylla. A high percentage of those observations were misdetermined. You might want to check to see whether the training set used for those species included some of the observations for which I corrected the determination. If it is trained using my new determinations, the CV should work a lot better to distinguish those two species.
See:
https://tchester.org/plants/analysis/collinsia/concolor_heterophylla.html
Yay! 2.14 is out! I’ve been waiting for a while :)
If every species included in this CV model would be a song/piece, you could play all of Beethoven's compositions 125 times.
By the way, the last paragraph starts with “We apologies”
@lj_lamera thanks, I fixed it.
I wonder if we, the observers can keep up with providing photos of "new" organisms. Actually I am now always happy if I find something that the CV doesn't recognise. Of course then we, the identifiers have to be able to give it a name.
Seen in 2009, uploaded this April, now in CV - thank you!
https://www.inaturalist.org/observations/208251089
@susanne-kasimir surely we do, there're thousands of thousands of species with 0 observations, and some extinct or rare species will never get through the treshold of CV model as it is now. Some groups that are ided by DNA only will never be actually correctly identifiable for the system, even if they're in the model.
Is there a list for species not included in the newest CV model? I'd love to help!
Speaking of that Oksanaetal, I'd also love to help if they need it.
For 'your chosen' species - click the About to see if it is Pending or Included.
We need about 60 obs to get 100 photos - then it will be included in the 'next' CV update.
If it is a taxon you know well, you may be able to retrieve the needed obs by going up taxon levels.
Thanks @dianastuder !
Sounds good, thanks for that information Diana.
This is tiwane's new Help for that topic
https://inaturalist.freshdesk.com/en/support/solutions/articles/151000170368-which-taxa-are-included-in-the-computer-vision-suggestions-
Gets the model much bigger and much slower by adding species ? If there are only 60 observations we do not get to the point that the price is bigger than the reward ? Or that specific models for birds/plants/countries are a better way to go ?
@ahospers I don’t believe the model is made significantly slower by adding more species. Even if it did, I think a model with as much species as possible is better than one with less, as long as it stays accurate.
Hooray! One of my favorite grasses, Sphenopholis interrupta, has been added with this release! I'll be looking forward to seeing how the CV model does with it next spring. I'd posted several observations this spring hoping that would help get it on the list. Exciting!
Hello,
Great work !
This species : https://www.inaturalist.org/taxa/484227-Carex-frigida seems to meet the requirement to be included in the computer vision model. Does anyone know why it is not ?
I like to revisit my old observations that are stuck at high taxonomic levels and check to see if the newest computer model can do a better job recognizing what I've observed than whatever model was in place when I first observed it.
@plantoine this model was trained on data from around the end of May. Most observations of it were added after that, so it might not have made the cutoff date for this version.
Awesome!
There's an interesting quirk with the map of observations of all the new bird species added. This isn't a "glitch" or a "bug", just the nature of the dataset: The map shows dozens of observations in North America (north of Mexico) and Western Europe of the newly added species, yet I'm pretty sure that no new species of native birds from those regions were a part of the additions. All of those observations are "Casual" because they represent captive individuals in zoos, etc., of species newly added from observations in their native ranges.
Agregar un comentario