Incorrect species and orientation in annotation (IoT)

Support
,
1

What Wildbook are you working in? IoT

What is the entire URL out of the browser, exactly where the error occurred?
https://iot.wildbook.org/encounters/encounter.jsp?number=c157e838-9e73-40e4-829e-d98aa374bfbd
https://iot.wildbook.org/encounters/encounter.jsp?number=8e98790b-e31b-4e9e-b921-4b675a4dee10
https://iot.wildbook.org/encounters/encounter.jsp?number=f88be2ac-2f81-4f74-9a8e-5892ca2423c8
https://iot.wildbook.org/encounters/encounter.jsp?number=275ab659-6a78-4f3b-964d-78295217ccca
https://iot.wildbook.org/encounters/encounter.jsp?number=8e1f0019-855a-4887-8c54-3c8f49e89e25
Can you describe what the issue is you’re experiencing?
After uploading, in the annotation the incorrect species is referenced. It also occurs there is an incorrect orientation.
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

2

Hello,

Thanks for providing all these examples. I am seeing what you mention with the species labels displayed on hover over the image annotation- some olive ridley and hawksbill labels appear for the green turtles you have submitted.

These labels are returned from the detection phase of our image analysis pipeline, but a human entered species will always take first priority so your images should match against appropriate candidates since you submitted them as Chelonia mydas. All the matching results I looked at for these images seem to be comparing against an appropriate set.

I personally annotated plenty of the data for the turtle detection models, and the vast majority of the on land images we had to work with were olive ridley and hawksbill turtles, so I am not incredibly surprised at the confusion. The model simply has not been trained on many examples of green turtles on land.

We went through this previously with hawksbills. On land and sand covered images were not behaving reliably, so a body of images was hand annotated and rolled into a retrained model. Tighter annotations, species recognition and viewpoints were produced with the new version.

I was only able to find one image in these encounters with an improper viewpoint, and it is definitely a mistake. I’ve manually corrected it and it should be matchable normally now.