"Hidden" Computer Vision API

On this forum, I have read and I was told that there is no open API to the CV model
(e.g., this answer; however, the same person also stated in the same thread that this is something they are talking and thinking about and the thread is already three years old, so I think it is reasonable to raise this question again).

Moreover, I have read in some paper (and also here in the Forum) that iNaturalist provides access to its Computer Vision model through API (apparently at https://rapidapi.com/inaturalist-inaturalist-default/api/visionapi/?) for a few “chosen ones”.

Now, I wonder whether and how it is possible to get approval to access this API.

Why would I need that access?
For research. I would like to include iNaturalist in a comparison of current CV models for plant species identification. iNaturalist is a well established platform with a huge community/user basis. Thus, it is reasonable to include the iNaturalist CV model in such a comparison, even though iNaturalist is not a specialist app for plant identification and even though CV identifications here are only a small part of the identification process which also includes community votes and discussion. But these are all considerations for a discussion part.
Uploading a number of images and writing down the suggestions manually is simply not something I could (or would be willing to) do, since it takes a lot of time, even for small data sets. (I am not planning to keep the iNaturalist API busy with thousands of requests per day, but even for lower two-digit numbers of requests per day, it is simply not manageable to make a standardised comparison of 4 to 6 CV models and write down results manually.)

The second question is usually: “Why would one need more studies if there are so many studies about the iNat CV model already?”

  1. We want to focus on a specific region. The performance of CV models varies considerably between regions of the world.
  2. Some of the studies published so far have been criticized (e.g., in this forum). Some used small test data sets. Many are not about plants. Finally, CV models evolve over time. If you want to know about the current state of CV, you don’t read a paper that is 3 or more years old.

So… what is the current state of the CV API policies? What is planned for the future?
And how and where would I apply to use the CV model?

was it really necessary to create this thread? you already have two other threads that cover basically the same subject:

seems like it would be better to just continue the discussion in an existing thread.

Well, I guess I did not define the issues clearly in each of these threads, so there is some overlap.
However, the core questions/issues are somewhat different, as I gained more information/a better understanding of how my questions can be misinterpreted and refined my questions accordingly.

further info from staff:


Thank you. Well, I have seen this post and sent an email. However, I did not get any response and, since this answer was also already two years old, how would I know if the author is still the contact person for these requests? It appears things have been developing at iNat and old forum posts are not updated, so I try to research a lot and my questions evolve with my knowledge about the topic… If the post contained a URL to a website that lays out the procedure and conditions on the use of the CV API, I would not have had to ask so many similar questions. Thus, the current thread could also be seen as a feature request. Maybe some webpage where things are clarified and information (such as conditions or contact persons) is updated…

iNat has a very small staff. And Carrie is (still) one of them.

1 Like

Thank you for that information. Then at least I know I might have to wait a few more days or hope she finds this thread (since emails some times end up in the spam folder). It is not a given that an email address from a forum post is still current.

as far as i can tell, folks understand your question just fine. it seems like you just didn’t like the answers you got. but creating a bunch of new threads until you get an answer you like seems a bit spammy to me.

if you want more help, maybe you should consider putting more work into showing folks what you intend to accomplish. if you’re comparing a bunch of computer vision models, do your comparisons on two or three of the other models first, and then come back with those to show folks what you’re working on.

i suspect that in the process of looking at those other models, you will learn the answers to a lot of the questions that you’ve been asking about here in the forum.

Staff also wrote somewhere that CV is developed to support the determination of photo’s added to iNaturalist. If you try to determine leaves and flowers separately or to determine a genus like Carex I think it will not work. Most photo’s are plants on location and are complete plants, not part of plants like leaves or flowers. And uploaded photo’s are often flowering plants, not stuff like grasses or sedges.

API-use is unequal app-use.

I did already test some of the other models. However, the policies and APIs are as different as it gets. Testing those other models did not really provide a lot of useful insights and my questions are very iNat-specific. Also, discussing further on these older threads is very unlikely to be helpful, since I would either have to edit the original posts (which would perhaps only create more confusion), or try to clarify it in the comments. In both cases, people will have to read through a lot of text until finding out what I really want to know.
The information on who is responsible kinda solved the issue for me, since I can now try to get in contact. Some official guidelines somewhere would probably still be helpful for people who want to find out in the future.

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.