Updated computer vision model and geomodel with over 1,500 new taxa
We released a computer vision and geomodel update today! Model v.2.25 has 109,680 taxa compared to 108,124 taxa in the last model and is trained off data exported on September 7, 2025.
This model is the result of millions of observations and identifications shared on iNaturalist. As new taxa are observed and identified, taxa (mostly species, but also genera, families, etc.) are added to the model once there are around 100 photos and 60 observations of that taxon.
As taxonomy changes and misidentifications get corrected by the community, sometimes taxa are also removed from the model. To keep up with these changes, we update the model every month or two so that the community can benefit from the improvements. Check out our help page for the most updated information on how we update the model.
Below are links to the taxa added since the last model. You can click through and search for your username to see if you have observed or identified any of these species. If so, thank you!
- Mammalia
- Arachnida
- Insecta
- Protozoa
- Amphibia
- Aves
- Fungi
- Actinopterygii
- Chromista
- Animalia
- Mollusca
- Reptilia
- Plantae
The graph below shows how the number of taxa included in the model has grown over the last few years — nearly doubling from 55,000 in 2022 to nearly 110,000 today.
Each time we release a new model, we evaluate it against the previous one.
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 2.24 with the new model 2.25.
Comments
Many thanks. Impressive as always.
As a matter of curiosity - if I check the link above to plants:
as per link above: 40,908 observations of 459 species
if I add the filter to southern africa I get 3,012 of 39 species
but if I try and filter for the genus Moraea it clashes with the &taxon_ids= and just shows all 43,875 observations of 198 species of Moraea.
Is there an easy way to filter out how many of these are say Iridaceae or Moraea, without manually counting them?
(There are 6 species in the genus Moraea in southern Africa in this run.)
Thanks for all the detail, it's always an interesting read.
Am I right in thinking that the last graph shows:
Of course, having more taxa in the model is great -- I'm curious how you think about the balance between accuracy and taxa coverage.
I'm sure y'all are already busy, but I was thinking it'd be neat to have a data visualisation tool that compares the CV accuracy between any two models. It'd be neat to see how much the accuracy has improved since I joined iNat in Jan 2024!
Thank you for another update! Are we getting closer to adding the ability for the community to directly determine what species are included or excluded from the model? There are still many plant hybrids which should be included in the model, and many genera and families which should be excluded to reduce the amount of CV slop IDs for taxa which are generally not identifiable from photos? There are several threads about this on the forum and it would be great for this to be added ASAP as these CV slop IDs are contributing to identifier burnout as we constantly fight the CV on taxa is performs poorly on.
I am curious: has anyone noted any species that the CV appears to "negatively ID"?
Two such species in the Proteaceae are https://www.inaturalist.org/taxa/592521-Protea-longifolia (1,142 observations) and https://www.inaturalist.org/taxa/577926-Protea-glabra (648 observations). These are readily identifiable species that are not easily confused in photos with other species. And yet the CV (v 224 anyway) usually identifies them as preferentially other Protea species (Pr glabra is usually in the list low down, but Pr longifolia is as often not even in the top suggestions).
I guess this is a far less problematic misidentification that falsely binning lots of species into one species. To me is is more amusing than a problem: the CV appears to have blind spots.
@kevinfaccenda also interested in this, because I notice a few fungi species that were added to the species list that are definitely not IDable based on macromorphology within their range - the only real reason they've gotten enough observations is because of sequencing efforts.
Crepidotus malachius and a few of the Russula are the biggest offenders that I see
@kevinfaccenda
I'm not sure anything is confirmed planned. If any progress on that has occurred, It has behind closed doors. The github repositories for CV related things have been pretty quiet for some time now. Staff have not commented directly on the feasibility of that proposal. So its really hard to say.
https://forum.inaturalist.org/t/force-computer-vision-to-back-off-on-the-specificity-of-suggested-ids-in-regions-with-cryptic-or-hard-to-identify-species/70021
When do you plan to add Messena sinuata (500+ obs.)???
Does this geomodel use GRID or SINR?
Messena sinuata was created on Oct 9, 2025 and this model is using data exported on September 7, 2025. But it should be in 2.26
2.26 uses GRID - we'll let people know if/when we switch back to SINR after sorting out those issues
Hi @aaravmishra. The species Messena sinuata was added to iNat (by you) on October 9, 2025. The CV model is updated roughly once a month, and, because of the time required for model training, the cut-off for observations tends to be about 6 weeks before the model is released. On that basis, you might expect M. sinuata to be included in the models released in November or December.
I understood. I were just excited.
Is it possible to see the number of CV species in different regions of the world? 110,000 taxa is a lot, but I want to know how many it knows where I live?
This is it for the region MONTENEGRO https://www.inaturalist.org/observations?verifiable=true&taxon_ids=63290,96844,103232,122022,129471,142214,195257,195875,210132,222760,294780,295356,326062,355862,361285,418080,426031,450023,496423,496790,543245,630795,671611,702494,712982,714951,732708,739134,739135,776071,782752,790986,851869,865381,940413,993450,1071981,1075493,1077434,1094712,1110863,1263140,1283826&view=species
ANd this it will be for Arizona https://www.inaturalist.org/observations?place_id=40&subview=map&taxon_ids=63290,96844,103232,122022,129471,142214,195257,195875,210132,222760,294780,295356,326062,355862,361285,418080,426031,450023,496423,496790,543245,630795,671611,702494,712982,714951,732708,739134,739135,776071,782752,790986,851869,865381,940413,993450,1071981,1075493,1077434,1094712,1110863,1263140,1283826
@ahospers - thank you. but I want to see ALL taxa included in the current CV...?
Add a Comment