Splitting head and body into two separate encounters

In which Wildbook did the issue occur? dont know what submitter used

What operating system were you using? (eg. MacOS 10.15.3) dont know what submitter used

What web browser were you using? (eg. Chrome 79) dont know what submitter used

What is your role on the site? (admin, researcher, etc) admin

What happened? We see ‘matched’ encounters where the head is in one encounter and the body in another, like it thinks they are two separate animals in one image. It duplicates the encounter and splits off the annotation.

What did you expect to happen? That head and body would be retained in a single encounter.

What are some steps we could take to reproduce the issue? No idea. But this may be what Chrissy thought/reported were duplicated encounters…? Example: head and body respectively

https://wildbook.seadragonsearch.org/encounters/encounter.jsp?number=9afc75c9-3c13-4a53-a60c-dcb8a6cd656e

https://wildbook.seadragonsearch.org/encounters/encounter.jsp?number=d24752d9-1dba-4e4a-b807-2253999ba935

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

Actually this also was subject to duplicated encounters, so we actually got two heads and two bodies show up after a single image encounter was submitted… So,

head 1: SeadragonSearch.org

Head 2: SeadragonSearch.org

body 1: SeadragonSearch.org

body 2: SeadragonSearch.org

Based on the occurrence ID, it looks like this was two separate submissions to which the same bug was applied: separating the body and head annotation.

Quick check: was this through normal submission and automatic detection or through manual annotation?

OK, I dug a little bit deeper here.

This looks like it happened through normal Encounter submission. I could repeat exactly what happened via that route. I did submit the photo myself, and only the expected two encounters were created.

First, I believe the submitter reported this Encounter twice. Across the four Encounters you list, there are two Sightings. Each Sighting (Occurrence) has two Encounters, each representing an Annotation found.

Second, the detector found two body annotations, not the expected body and head. Thus the Encounter was split into two under one Sighting. Here’s what the detection result looks like under the covers for this photo.

{
‘response’: {
‘jobid’: 8f44dfec-3022-4c22-923d-2a0722cc3514,
‘json_result’: {
‘image_uuid_list’: [
8d9db276-2382-1c5f-d01c-6cbecfeac3e4,
],
‘results_list’: [
[
{
‘class’: seadragon_weedy,
‘confidence’: 0.9076,
‘height’: 3086,
‘id’: 13742,
‘interest’: False,
‘left’: 290,
‘multiple’: False,
‘quality’: None,
‘species’: seadragon_weedy,
‘theta’: 0.0,
‘top’: 808,
‘uuid’: 28b679a1-9110-4b74-8a0b-7d082242b818,
‘viewpoint’: frontleft,
‘width’: 5503,
‘xtl’: 290,
‘ytl’: 808,
},
{
‘class’: seadragon_weedy,
‘confidence’: 0.8461,
‘height’: 1864,
‘id’: 13741,
‘interest’: False,
‘left’: 521,
‘multiple’: False,
‘quality’: None,
‘species’: seadragon_weedy,
‘theta’: 0.0,
‘top’: 501,
‘uuid’: dc648a9f-47d1-4680-a988-63d26cffb3f6,
‘viewpoint’: frontleft,
‘width’: 3695,
‘xtl’: 521,
‘ytl’: 501,
},
],
],
‘score_list’: [
0.0,
],
},
‘status’: completed,
},

All in all, this looks like a double submission by the submitter and a misclassification by the detector. While the misclassification is unfortunate, they happen a % of the time. This does look like the system working as designed. It can be fixed easily by deleting the extraneous Encounters manually and by manually annotating an additional head.

Thanks,
Jason

Hi Jason, when a manual annotation is created now, it goes to a separate cloned encounter. But its hard to keep track of two separate encounters for the one animal. It used to attach to the same encounter…any ideas how to merge encounters back together again? thanks, nerida

Hi Nerida,

In the case of seadragons, the head is a “part” and shouldn’t cause an automatic Encounter split. You have a good point.

I have filed ticket WB-1118 and have a test fix on your Wildbook. If you add a second body annotation, a new Encounter will be split off. But if the annotation added is a part (e.g., seadragon_leafy+head), it should no longer split the new annotation to a new Encounter but rather keep it on the same Encounter.

I’m hoping this addresses your issue. Can you please give it a try?

Thanks,
Jason

Thanks Jason, it has now created a head annotation attached to the same encounter! However, it appears offset, which I reported in a separate thread.
https://wildbook.seadragonsearch.org/encounters/encounter.jsp?number=a69e231f-e8c2-497b-908d-5cd978683d31

Cool. We’re tracking that as WB-1101. Looking at it further, it looks like it only affects these narrow, vertical images. Likely that makes it a small math bug.

When I rotated the vertical image to the horizontal before uploading - as a workaround - my manual bounding box appeared correctly.

I wondered about that! Thanks Jason. Thats is a totally doable workaround for us.

Hi again Jason, I used this workaround for the problematic encounter. However, now it wont allow me to do a search or match of any kind on the new manual annotations.
thanks for chekcing
https://wildbook.seadragonsearch.org/encounters/encounter.jsp?number=a69e231f-e8c2-497b-908d-5cd978683d31

I think maybe we have been running a lot of tests on this same data. :slight_smile:

Two things happened.

  1. Taxonomy is not set on this Encounter, so any uploaded photo can’t go to detection and get a detectionStatus=complete, which is a precursor for an annotation to be matchable.

  1. Without that detectionStatus on the Annotation and without a Taxonomy, the system doesn’t know how to match it as our matching pipeline is based on species.

I recommend deleting this Encounter and resubmitting from scratch, setting the species on the new submission.

Thanks,
Jason