Can you describe what the issue is you’re experiencing?
I uploaded these two encounters to see how well the platform would do with annotating the fish and it really hasn’t annotated any. It seems to be working worse than it had in the past. Particularly for the photo with the fish close up.
Can you provide steps on how to reproduce what you’re experiencing?
Uploaded encounter and then looked to see what was annotated.
If this is a bulk import report, send the spreadsheet to services@wildme.org with the email subject line matching your bug report
It’s interesting that in the first link, it found the fish in the shadows but not the ones in the foreground. This one would probably be best to manually annotate.
For the second encounter, no species was set so detection hasn’t run. An image like this would likely create hundreds of encounters for each detected fish if fully annotated.
Detection is typically pretty accurate, but it does miss the mark sometimes. How often are you seeing missed annotations? Do you have other examples you can share?
Hi. I had actually uploaded these two images to check and see how well it would annotate them because we are working on a boxing algorithm for video processing.
Can you explain why it missed the foreground images? Was it because they were too close?
I will keep an eye on other images. We generally upload cropped images of one fish so I haven’t paid that much attention to it
Typically, this sort of thing happens for different reasons, including but not limited to:
the background being too similar to the foreground
the image is of a viewpoint that was underrepresented in the training data
the image isn’t representative of the images the detector was trained on (such as lighting conditions, lab environment vs natural habitat, etc.)
the image is fine, the detector just dropped the ball without a discernable reason
Detection missing the two examples you gave aren’t giving us a huge cause for concern, but if you do have other examples for us to look at, please let us know.