I do not think identifiers should use CV at all.
A somewhat awkward work-around is to use the CV Demo instead. Most browsers have a screenshot tool (or if not, an add-on) that allows saving a region of a web-page to disk. These tools often automatically pre-select any images on the page, and you can crop the area containing the subject, which often greatly improves the CV results. I keep the CV Demo open in a pinned tab, and have a button for the screenshot tool on the main toolbar. (NB: refreshing the demo page restores its āChoose Photoā button). Itās not exactly user-friendly, but better than nothingā¦
There are several legitimate reasons for using the CV whilst identifying. Probably the most common one for me is using it as search tool when I know exactly what something is, but cannot remember its name. I also occasionally use it for sanity-checking when Iām only 90% certain of my own ID. And of course, sometimes Iām just curious to know what the CV makes of some photos.
10 Likes
I was going to list the same things. Itās a tool. Like any tool, it can be helpful if used properly.
12 Likes
another ironic reason is sometimes the CV suggestions gets me faster IDs than typing and waiting for autocomplete result and also than picking right ID upon collisions of similar names suggestions. sometimes :)
i do wonder of that since CV should be computationally demanding than simple autocomplete.
3 Likes
CV is more efficient. Autocomplete struggles even with @mention the observer.
2 Likes
You (all) maybe do not understand me.
I am using CV to autocomplete the IDs of my own observations all the time! Most of my observations will have the CV symbol even for taxons that I know very well.
However, I do not believe that IDers should use CV when identifying observations of other observers in the Identify tab.
1 Like
What is the difference between case 1 and case 2? Identifiers may be using the CV for other peopleās observations for exactly the same reasons that you use it for your own observations. This does not mean they are thoughtlessly entering the first CV suggestion any more than you are when you use the CV as a shortcut to get the taxon you want.
I do a fair amount of broad IDing of arthropods. Sometimes if I am mystified I look at the CV suggestions to get ideas about what it could be ā as a starting point for my own research. A lot of the time, I do not end up entering the CV suggestion, but some broader taxon. If I have no idea what something is, it is difficult to know where to start in trying to figure it out. The CV provides possible avenues for investigation, which I then supplement with my own knowledge. This is not ācheatingā any more than consulting a reference book would be. Just because someone is using Identify doesnāt mean that they are required to know everything they identify so well that they do not need any other resources. Using Identify also does not mean that we are expected to have perfect memories; not being able to immediately remember the name of an organism does not mean we do not know how to identify it.
6 Likes
āComputer Visionā and āAutocompletionā appear to be two separate functions on iNaturalist, guessing from their API documentation. One is enforced on all uploads and ID form fields (no way to turn it off preemptively) ā and commonly used by identifiers as a sluggish form of autocomplete; the other is not enabled independently anywhere, as far as I can tell.
edit: in a parallel universe, iNat could enable CV (and the Agree button as well, but thatās for a different thread) for the original uploader of the obs only, precluding its later use by identifiers. While enabling autocompletion of inputs, to keep identifiers happy.
1 Like
Indeed, it is different. I just let the CV to display its suggestions on my observations and choose the one I want, if it is present.
1 Like
And why do you think identifiers should have to type out their IDs just because they are identifying for others instead of IDing their own observations?
Um, in this universe the CV suggestions are already not activated if one is typing in oneās ID from the āinfoā tab of the Identify module. To access CV suggestions one has to view the āsuggestionsā tab, where it works very differently than the CV on observation and upload pages.
Why do you think identifiers should be āprecludedā from using the CV as a shortcut for getting the ID that they want? Having the option to use a tool does not mean required to use it, nor does it mean that people who use it are doing so irresponsibly.
RG observations should really better be at least one user using what they know and the owner pressing Agree than just one user using CV and the other blindly pressing Agree.
2 Likes
I agree. But just because someone uses the CV does not mean they are thoughtlessly selecting the ID ā you say you do not do so for your own observations.
You have not answered my question: Why do you believe that IDers are thoughtlessly using the CV for other peopleās observations rather than using it as a shortcut to get the ID that they want?
Many observers uncritically apply the CV to their own observations. But in my experience most people who take the time to provide IDs for others are doing so with some knowledge and are doing so responsibly.
The use of the āagreeā button is another issue altogether ā and it is available to both identifiers and the original observer. In other words, if people are going to irresponsibly confirm IDs, they do not need the CV to do so.
6 Likes
To try and keep to a strict minimum these cases whereby an identifier does not know what they want, then defer to a black boxās decision? The layout of the āIdentify moduleā was indeed a step in the right direction. (Letās not derail the Request thread; discussing the pros/cons of machine interference with human intellectual processes is too heated and off the scope of iNat or this forum.)
Well, you started it.
And yet there is a difference between making an option available and requiring people to use it. Users are never required to use the CV suggestions in any of iNatās interfaces even if there is no way to prevent the suggestions from being displayed in some contexts. (If you wish to submit a feature request for a setting that would allow users to turn it off if they prefer not to use it, I would support this.)
There is also a difference between discouraging the use of an option and preventing people from using it at all.
The vast majority of the wrong CV IDs that I see are those made by observers, not those made by identifiers.
The vast majority of wrong RG observations that I see are the result of observers uncritically clicking āagreeā on an erroneous ID. Some portion of these wrong IDs may have originally been CV suggestions; a meaningful percentage are the result of wrong selection from the autocomplete dropdown menu (something you are in favor of); others are simple error that have nothing to do with tools made available from iNat. There is also a small portion of cases where wrong RG observations are the result of the observer using the CV and a friend clicking āagreeā, so here again the RG status is not the result of identifiers using the CV.
Conclusion: Removing the CV will probably not meaningfully reduce the number of wrong identifications made by IDers.
4 Likes
Huh, nope. There is no autocomplete (dropdown menu or not) on iNat at the moment. Only CV. And at the moment, in this universe, there is (understandably) zero discouraging its use, either by explicit guidelines or by technical means.
IIRC it has already been suggested and (quite understandably) dismissed.
This is either obviously false, or you havenāt explained yourself clearly.
The ID input box supports both CV suggestions and autocompletion. The former only works when the input box is empty. However, as the name implies, the latter requires you to type something in first, otherwise thereās nothing to complete. The resulting drop-down lists are not the same.
4 Likes
On the ā/observations/#ā pages there is no way to use the āSpecies nameā form field without having first a ānudgeā by the CV, then, upon hitting some keys, a second different ānudgeā by whatever algorithm orders the new suggestions in some unpredictable way (itās definitely not alphabetical, and does not disclose how it chooses and why: geography? color? string length? random?).
On the ā/observations/identifyā pages there is no way to use the āSpecies nameā form field without having a ānudgeā by whatever algorithm orders the suggestions in some unpredictable way (itās definitely not alphabetical, and does not disclose how it chooses and why: geography? color? string length? random?).
Moved the above discussion from https://forum.inaturalist.org/t/use-computer-vision-on-each-photo-in-an-observation/4210/
Maybe technically youāll see CV suggestions when clicking in and before you start typing (depending on internet connection, etc) but I donāt think thatās meaningful. If you donāt want to use CV, just type in what you think it is.
3 Likes
So, you figure that you can use the CV responsibly but I cannot?
12 Likes
If thatās the case, then do you think they should not use the āAgreeā button either?
1 Like