Curious for advice about making the first specimen photo a diagnostic photo even if that photo is of a microscopic feature? I do a lot of seaweed observations and require students to take multiple photos — at least one field shot showing the substrate, a pre-dried specimen on herbarium paper to show gross morphology and any additional photos that show diagnostic features needed to key out the seaweed.
My preference would be to have a key diagnostic photo first, even if it doesn’t show the whole organism. The reason for this is that it is often hard to know if a record has multiple photos when viewing as a list of observations (for a site, project, etc). Thoughts?
4 Likes
You can order the photos any way you want. Having a diagnostic photo first would save time for identifiers, so that is what I would go for too. A shot of the entire organism would also work for the first photo.
10 Likes
I would agree.
I would go even further in a sense and say the main taxon image should be whatever is diagnostic in these circumstances… e.g. with flies where you need an image of their genitalia to ID, like Sarcophaga carnaria, that should be the title image… I think otherwise it misleads users… seeing an image of the fly itself as the taxon image encourages superficial comparison ( especially if using the compare tool )
Also for the AI model - the training data prioritises usage of the first image.
6 Likes
For working entomologists - but most iNat obs will not show that.
The first taxon image should show field marks - but at the level of what we can hope to see in ‘average’ photo. Which might mean the ID goes to a higher taxon level, with a note from a scientist cannot ID to sp without … On second thoughts, if the first taxon picture is obviously from a microscope, perhaps that is an effective solution.
I am wary of those obs, as they can be for multiple species, and need to be split before we add an ID
Only one species per observation please
https://help.inaturalist.org/en/support/solutions/articles/151000171680-what-do-i-do-if-the-observation-has-multiple-photos-depicting-different-species-
Please don’t ID for only ONE of the multiple species - until the observer has split to single species.
new DQA Feb 2024 - evidence related to a single subject
I am not a working entomologist, but my observations for this species detail this.
(and have them as the initial image)
Not sure what you mean. iNat obs which don´t show this simply can´t go to species in this instance. So for Sarcophaga carnaria, all existing iNat obs should show this.
1 Like
Most iNat obs of flies are live - is what I meant.
I defer to your knowledge of flies.
Ah I see, yes sorry, gotcha - I misread :)
1 Like
Oof… when did it change?
the last time I checked, ordering did not matter to the CV:
https://forum.inaturalist.org/t/what-image-s-are-used-for-training-computer-vision/3307/2
1 Like
It’s also my understanding that that CV training set can pull from any image in an observation.
The CV will use the first photo in a multi-photo observation for providing suggestions for a given observation.
3 Likes
Oh interesting.
I thought I saw @tiwane or @alex mention that at some point in one of the more recent updates.
But maybe I am misremembering then!
I guess I would presume it would make sense to place a bit more weighting on first photos at least… as latter photos are more likely to include more vague habitat shots, etc.
But yes, maybe I´m wrong in these presumptions then haha :)
I agree with responses that it’s a matter of personal choice. I’ll admit I usually make the first photo the one I aesthetically like the most. After that I usually prioritize whole organism photos before single organ photos before alternate angles of the aforementioned, and whatever else.
4 Likes
Personally, I would always want the first photo to either be the best general photo, or if applicable a photo with a relavant particular feature, just so that a time-pressed identifier doesn’t just think ‘Oh another blurry one’, and move on without looking at the rest. Just something that’s a bit of a hook.
7 Likes
And field mark instead of eye candy?
2 Likes
Yes, indeed. Straight away gives the signal ‘one that’s got a good chance of being identified.’
3 Likes
I’m not a working entomologist, but I do need those features when I am identifying. Also, not only entomologists or avid insect enthusiasts need photos w/ microscopy such as those of the genitalia, with spiders that is very often needed as well.
1 Like
My thought process on this is, it doesn’t really matter considerably. Putting the image w/ the most fields marks/most important feature(s) is the most helpful for identifiers - to an extent. What I’ve found is that sometimes even if you upload a photo of, say the genitalia or other mark which needs microscopy to identify sometimes it will just be ignored. What I do and what I would recommend is posting first the photo that covers the most of the specimen (e.g. a full and detailed body shot). Next, include different angles and specific identification features. Both are fine and both have their drawbacks.
Again, you can honestly do it anyway you want so it doesn’t matter all that much so do it the way you prefer most!
We are talking at cross purposes.
Identifier needs more info - happens for plants too.
Taxon specialists see a different slice of iNat (I have never seen microscope pictures on a spider obs - which doesn’t mean they are not on obs I don’t see because they are already RG)
What do you mean by that?
I mean that we agree.
Need more information for an ID.
That’s an interesting hypothesis.
One roundabout way of thinking about this that wouldn’t require re-training a full model just to explore the idea: if I take a batch of multi-photo observations and only do suggestions on the second photo, do we get worse results than we normally would for those observations? If the model performs worse on second photos than it does on first photos, then that might suggest that they contain different content.
2 Likes