Vertical pictures misplacing annotations

Thank you for the clarification!

I ran into an issue. One of my images, a vertical one, keeps placing the annotation box incorrectly.
https://iot.wildbook.org/encounters/encounter.jsp?number=0cee16d7-b586-4072-a1ce-52c1a3a223f3
And after manually annotating, it did a similar thing
https://iot.wildbook.org/encounters/encounter.jsp?number=067adf9d-aab3-430b-a1b1-4b8f823e29d3

same thing happened on this one, again on vertical shots
https://iot.wildbook.org/encounters/encounter.jsp?number=8821e801-8967-4431-860a-c9d105a3ddfa
https://iot.wildbook.org/encounters/encounter.jsp?number=217b831e-5400-4b1e-acbd-61fb900738bc

Any thoughts on what is going on?

Hey @noaahawaii,
I pulled this into a new thread since this requires a new round of investigation. Will have someone start looking into it right away.

Tanya

@noaahawaii,
We’ve done some digging and got some fixes in for the drift happening on the automatic annotations. The first encounter link you posted seems to have been fixed with these changes. The second link, the manual annotation, I’m pretty sure you can delete now.

Hello @noaahawaii,

We’ve just deployed a couple of changes to IOT and other platforms that should help with the issues
you were experiencing.

First, bounding box placement should now come out as expected for horizontal images. Sometimes orientation information drawn from certain camera’s and phone’s was causing display issues. Please let us know if you see this again.

Secondly, when you delete an encounter it now removes any extra manually drawn bounding boxes
associated with it. I see the a couple of the encounters you were having trouble with retained extra
boxes associated with no encounter. This should not happen in the future.

Hope this helps. Thanks for reporting these.

1 Like

Do I have to re-run/ start another match on these images I uploaded? I noticed on some that once you ‘Inspect’ to compare it does not read the correct part and compares a section of sand for instance. Or is there a way to re-run my last 50 uploads for instance?

Thank you for checking these issues so quickly

If the incorrectly placed bounding boxes are manual annotations that didn’t end up being placed correctly due to the image orientation error, unfortunately there is no automated way for us to fix them.

They must be removed by deleting the encounter associated with the manually added bounding box, or at least a correct one must be manually added.

Incorrect boxes left on encounter could also be left alone while new ones are added. Since they don’t enclose complete matching regions it will not likely interfere with identification. Sorry about this extra labor.

1 Like

Work done to resolve these issues was tracked in the following tickets:
WB-717 and WB-794

Hi! I went back to this encounter and it is still giving me issues with the annotation (even a manually one) drifting to the incorrect location. I have deleted the encounter and uploaded it again, but nothing seems to change

https://iot.wildbook.org/encounters/encounter.jsp?number=adc7568a-4b97-4c67-8202-e120132574ab
https://iot.wildbook.org/iaResults.jsp?taskId=b483aba0-6add-4c1d-b3f8-5e1216ce2fe5
https://iot.wildbook.org/iaResults.jsp?taskId=82418227-0962-43fa-9594-0c35265e40a1
It is also comparing left side to right side annotations occasionally.

Hello,

Thanks for providing the examples for this.

I’ve taken a look at this particular image and it looks like orientation EXIF data saved by the camera or another utility has inaccurate rotation information. The camera is noted as a Panasonic DMC-FZ28 in the EXIF data, but I’m not sure what settings determined this saved orientation.

This leads to the image displaying okay in the browser, but when we try to save bounding boxes the system respects the orientation given by the image and they are saved out of place.

I’ve stripped EXIF from this image and submitted it again, and the annotations appear correctly:
https://iot.wildbook.org/encounters/encounter.jsp?number=8a051d89-f4b2-45fa-afea-8ac5db38f557

Here is the online utility I used to do this:
https://www.verexif.com/en/

If you strip this data for an image with this behavior it should show up as you expect. This will apply to human added bounding boxes as well.

Once you have resubmitted this image I’ll delete my test record the the image is not saved twice in the database.

Thanks!

1 Like