iNaturalist - A New Vision Model

[jch2103]

I participate in inaturalist.org, a (free!) worldwide citizen science site that collects observations (photos, text, locations, etc.) of animals, birds, insects, plants, etc., etc. One of the features of iNaturalist is a computer vision model that it uses to make automated identification suggestions (something like face recognition...). In March, they introduced a new vision model for automated identification. I've noticed that the suggestions keep improving, even for photos where the subject isn't dominant.

Those interested in the technical background of the IMatch face recognition system may find a description of the iNaturalist system informative: https://www.inaturalist.org/blog/31806-a-new-vision-model

I recommend iNaturalist to anyone who photographs critters, plants, etc., and who wants to contribute to a global community of citizen scientists.

[Aubrey]

I use Cornell labs app for bird identification
https://www.allaboutbirds.org/guide/search#
It's pretty good, but not great for migratory birds on Cyprus

[Menace]

Another great tools:

- Flora incognita (unfortunately just as an App yet) -> for Germany
- https://observation.org  (upload with great indentification) -> for middle Europe

[sinus]

A lot of interestic links, thank you all.

[Aubrey]

Another great tools:

- https://observation.org  (upload with great indentification) -> for middle Europe

I must check this out.
Thanks.

[Mario]

The offer even an API (REST endpoints): https://api.inaturalist.org/v1/docs/ which makes their features available for anything that can access REST endpoints...like IMatch apps... 

[Jingo]

another app (android/iphone) that I use and recommend is PlantNet... snap a photo on your phone of plant (leaves, flowers, fruit), click the button and voila!  It actually has been very very accurate for me and helped identify some plants on my property that I have been misidentifying for years.... such as wild rose which I always thought was blackberry (and wondered why it never produced fruit!).. 

[Mario]

They also have an API, which could integrate this info IMatch... 
https://my.plantnet.org/
50 free identifications per day, 1,000 identifications for 3 euros.

These services would probably fit nicely into the IMatch AutoTagger, as a new service (in addition to Google, Microsoft, Clarifai and imagga).
AutoTagger has all the plumbing required, just a new interface which transforms the results into headlines, descriptions, keywords...

Not sure how much demand there is for this, though...

[Menace]

We just discuss this last year and now, it almost there.

[Mario]

Integrating these (or similar) services into the AutoTagger will be work, but doable.

I have no idea if there is demand for such an integration or how many users would use it.
I count 3 users in this post...

Less than 5% of the user base has used the AutoTagger in the past 3 months.

Not sure if there is no demand.
Or if the people have issues with privacy.
Or if the setup with Google/Ms is too complex.
Or if people are not willing to pay for these services.
...

This would be something for a poll or an email questionnaire...

[Menace]

There are more points: How is the integration for report the species in correct terms. A lot of people, who use these APPs and webpage are contributing the information for science. If this can't handled well by the Api they would switch back to the Original-Website.

Another point would be: If someone use multible these Apis, he could get (maybe, not sure of this) better results.
Or what if in one year is a much more powerful API from another company/university online?

So, I think a poll would be the first thing, which should be done.

[mastodon]

I think these autotagger give too many keywords for a home user, but not enough (and not specific). For corporate ones they are more useful.

[Mario]

So, I think a poll would be the first thing, which should be done.

Unfortunately, like with telemetry, not many IMatch users participate. In the community. Or polls.
And 'mass' is what counts here, unfortunately.

[Mario]

I think these autotagger give too many keywords for a home user, but not enough (and not specific). For corporate ones they are more useful.

AutoTagger has been included mostly for commercial users. And for users who start with DAM and have to deal with 50K or 500K of completely untragged files.
If you have to add keywords to 100,000 files, you willingly pay a small sum for automatic tagging and to do it in a day instead of several weeks. Even if some of the keywords are wrong or missing.

If you 'only' have to tag your own photos, as the come in (e.g., after a shooting or a vacation), tagging is not that much of a problem.

There are also some add-ons for AutoTagger for commercial clients, e.g. for identifying athlete numbers when covering sport events.
Add-ons for identifying plants or bugs or animals are thus thinkable. If there is sufficient demand.

[jch2103]

They also have an API, which could integrate this info IMatch... 
https://my.plantnet.org/
50 free identifications per day, 1,000 identifications for 3 euros.

These services would probably fit nicely into the IMatch AutoTagger, as a new service (in addition to Google, Microsoft, Clarifai and imagga).
AutoTagger has all the plumbing required, just a new interface which transforms the results into headlines, descriptions, keywords...

Not sure how much demand there is for this, though...

Interesting. But note that the API currently just supports plant identification. I do sometimes have iNaturalist help ID wildflowers, but for the most part I use it to ID critters (e.g., birds and dragonflies/damselflies).

[Jingo]

I think these autotagger give too many keywords for a home user, but not enough (and not specific). For corporate ones they are more useful.

That has been my experience so far... even turning down the number of returned results can still provide too many hits that I do not need.  I found myself just adding keywords because they were returned... but then never searching on those keywords to find items in the database.  Great to have the option though and the fact that other API's are available gives us APP coders something to do.. :-)

[DigPeter]

I word of caution is needed over these species identification apps.  They are not foolproof.  I am a botanist and have tried a few.  I would say at best they are about 80% accurate.  I would not wish to use the results to tag my photos.

[Menace]

The webpages have mostly a identification by other users. Flora incognita is over 80%; but they show like other apps their level of accurateness. 

[Mario]

I think these autotagger give too many keywords for a home user, but not enough (and not specific). For corporate ones they are more useful.

That has been my experience so far... even turning down the number of returned results can still provide too many hits that I do not need.  I found myself just adding keywords because they were returned... but then never searching on those keywords to find items in the database.  Great to have the option though and the fact that other API's are available gives us APP coders something to do.. :-)

Since all services have a rather limited vocabulary you can usually quickly suppress unwanted keywords in AutoTagger. You can also control how many keywords AT requests to limit it to the 5 or 10 best. The default is 20. I shall set the minimum to 5 instead of 10. Having few extra keywords does not hurt, even if you are not a stock photographer