Taking a *lot* of photos to assist CV?

A found a few shells of a species that has ~10 observations on iNat and around 30 photos. Species tend to get added to CV and the geomodel at around 100 photos. Would it be a good idea to take a lot of photos (more than 10) with a lot of different backgrounds and angles to add to a single observation, with the intent to speed up the process of getting it into the CV? Are there any downsides to this?

Also, here’s an example of a few different photos I’ve taken of the same few shells. I haven’t uploaded them yet.

6 Likes

Apparently only up to 5 photos from an observation might be selected for training (could be fewer than 5 if randomized), so you’d need to observe many different individuals of the species.

If your photos represent most of the observations it’s trained on, it would then be optimized to your camera settings and photo choices rather than those of a bunch of different people. So that could be a downside compared to a more “natural” progression of observations being added to iNat.

11 Likes

iNat chooses photos from various photographers.
So ‘one person cannot add 100 photos’ for CV.

2 Likes

That’s not correct. Here is an example of a species familiar to CV with 50 observations out of 64 made by one observer: https://www.inaturalist.org/taxa/868711-Megadenia-pygmaea

2 Likes

Cannot get the link to open. Internet and iNat are VERY slow tonite.
https://inaturalist.freshdesk.com/en/support/solutions/articles/151000170368-which-taxa-are-included-in-the-computer-vision-suggestions-

I did this for 4 small marine bivalves: the Lunate crassinella, three-tooth carditid, flat cardita, and grey pygmy venus. I took 4 photos each of about 14 shells from each species (front & back on two different backgrounds) which is about 50 new pictures per species. The first two are now included in the CV model!
for example
https://www.inaturalist.org/observations/168110734 (grey pygmy venus)
https://www.inaturalist.org/observations/168123150 (flat carditid)

4 Likes

I’m hoping to get Asphondylia lacinariae (Liatris gall midge) into the CV set. I took pictures of multiple individual galls from multiple angles. There is a lot of variability in the shape. But then I realized that all the ones I photographed are on the same host plant, Liatris pycnostachya var. lasiophylla, which would effect the training. This year I’m going to try to find some on a different species of Liatris.

5 Likes

I tried to vary things like how zoomed in the picture is, the color and texture of the background (I was biased towards sand background and hand shots as that’s how the majority of seashells are posted) and I might have also messed with brightness and saturation but I don’t remember.

I think I originally talked about it here
https://forum.inaturalist.org/t/how-are-photos-selected-for-cv-training/42403

2 Likes

Five years ago when I was really into galls, particularly of hackberries, I tried to get as many species above 100 as I could. Here’s an example for Hackberry Barrel Gall Midge: https://www.inaturalist.org/observations?place_id=97394&preferred_place_id=1&subview=map&taxon_id=947783&view=observers

At the current tally, I have 100 out of 116 observations. At the time I did them it pushed it above the threshold for inclusion. Five years later, if it were not for them, it would still not be close to inclusion. I’d say it’s a very good thing to do.

5 Likes