Display photos for suggested IDs that match observation's life stage

I uploaded an observation of a larval-stage moth, but all the suggested IDs in the dropdown displayed photos of species’ adult stages. Given how different the stages appear visually, photos that match my observation’s stage would be more readily helpful in making an ID.

Thanks!

The AI has not been trained on life stage data, so this wouldn’t be possible at the moment.

Tapping on the “i” button on a suggestion will allow you to see photos of the taxon that have been chosen by many users. For lepidoptera, most of them show the larva as the 2nd or 3rd photo.

1 Like

Thanks for your reply, tiwane.

Is there a plan to train on life stage data?

If I annotate my observation’s life stage, can the drop down menu show photos from observations where someone else has manually annotated the life stage as the same as my observation (i.e. if I label mine larval, show me photos user-tagged as larval). That wouldn’t require any data re-training and would nice to have.

It’s certainly something we’ve discussed and would like to do, but it is not a priority, considering our limited resources and the time it takes to train a model (right now it takes a few months with our current set up).

No, but on the website you can choose to only look at photos annotated as a certain life stage. eg https://www.inaturalist.org/taxa/127133-Hyphantria-cunea/browse_photos?term_id=1&term_value_id=6

2 Likes

Have to confess I’d vote to put the limites resources on adding geographic criteria to the AI IDs over life stage :-/

5 Likes

I’m not quite sure how this would work technically. The insect life stage annotation only appears for entry after an id of an insect has been entered.

You’d basically have to enter an id without the benefit of it, go enter the larva or whatever lifestage, and then go back and rerun the computer vision suggestion a second time.

For a point of clarification, pics of larva etc are not excluded from the teaching model, they are in the set for any species. In fact, the tool seems to do a very good job (at least for the Canadian stuff I observe) if presented a photo of a caterpillar etc of correctly identifying it.

2 Likes

I’m going to close this request as it’s not something we’ll be moving forward with. Training the data won’t happen any time soon (if ever) and the images can be found on taxon pages.