Annotating and detection of sea turtle head photos

Support
,
1

What Wildbook are you working in? Internet of Turtles

What is the entire URL out of the browser, exactly where the error occurred?

Can you describe what the issue is you’re experiencing?
We have a large database of marked individual turtles from our fieldwork that we are working on loading into IoT. We’ve intentionally prepped photos of the heads of the turtles from 3 different view points to populate this project with “marked individuals” in hopes of allowing matches from other users. The photos are typically of the head from the right, left, and down (top) view; and we’ve cropped them to make viewing easy.
However, the images do not get automatically annotated correctly - and we’ve only been able to get it to work (find matches) when we manually annotate, and do the visual match. I then select the view and the “green_tutle+head” option, and then draw the outline box and save.
Wondering if IoT is simply not used to these types of photos - and maybe we could help train? Open to any ideas, as the detection (mark annotation) and match process is currently very slow and time consuming. Thanks for any thoughts!
See examples here:
(after manual annotation on all 3 photos for 2 different animals): Internet of Turtles
Internet of Turtles

Hoping that getting this step working better/easier will then help the matching process improve.

Can you provide steps on how to reproduce what you’re experiencing?

If this is a bulk import report, send the spreadsheet to services@wildme.org with the email subject line matching your bug report

Many thanks!
C

2

Hi @CTurner

Thanks for sending over your spreadsheet!

Our algorithms have primarily been trained on left and right sides of the head for head annotations. You may have less luck matching top-down views. But the fact that you have all sides cataloged for your individuals is helpful because those viewpoints will now have IDs associated with them for future training efforts.

As far as detection not finding the turtles in your photos, my best guess is that they’re too closely cropped. According to our photography guidelines, while you want the animal in focus and prominently in the image, when the full animal is too close to the borders it can be harder for the detector to know what to do with it. You may want to try uploading a test encounter with the uncropped version images to see if the detector works better on those.

The only callout I have about the spreadsheet is that you’ll want to include the file extension in the media asset name (such as turtle1.jpg vs turtle1) for future uploads. I’m surprised it wasn’t flagged during validation, but if it worked this time, we’ll leave it alone. :sweat_smile: