Amphibian and Reptile: Bulk import Detection Error

[Mirjam]

In which Wildbook did the issue occur?

Amphibian and Reptile

What operating system were you using? (eg. MacOS 10.15.3)

Windows 11 Business

What web browser were you using? (eg. Chrome 79)

Microsoft Edge for Business (Version 152.0.4191.53 (Offizielles Build) (64-Bit))

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

researcher

What happened?

I uploaded three bulk imports. I did not skip detection or identification. The import worked for all bulk imports. I added the new photos to their respective projects.

Among others the spreadsheet included the following columns: ‘Encounter.genus’ (‘Bombina’) and ‘Encounter.specificEpithet’ (‘variegata’) so wildbook should have the information of the species.

There were no errors or missing fields in the spreadsheet, and the species was filled in correctly on the ‘Check the spreadsheet’ webpage: Bombina variegata.

On the bulk import progress webpage, ‘detection’ was marked as complete, but no ‘class’ had been added to the photos. This problem occurred with all three bulk imports.

Furthermore, when viewing the new photos on the project webpage, the detection status was ‘error’. No annotation had been added to the photos.

Identification did not start automatically, and I cannot start it manually either.

For the bulk import that only included two photos, I added the annotation manually, after which the identification process began. However, the other bulk imports are larger, so I would prefer not to add the annotations manually. As far as I can tell, the photos are OK, so Wildbook should be able to add the annotations.

What did you expect to happen?

I expected the annotations to be added automatically, after which the identification process would begin.

What are some steps we could take to reproduce the issue?

Upload a bulk import.

Thank you in advance for your help!

[Anastasia]

Thanks for sending your spreadsheet! I’ll need a little more time to dig into this as I’m still seeing the detection errors when I re-run it.

[Anastasia]

Thanks for your patience. Viewpoints weren’t configured for this species. I’ve resent your import to detection and I see annotations now. Some may need to be fixed manually as this type of viewpoint wasn’t well represented in the training data, so the automated detector may make some mistakes. I do see more good annotations than bad ones, though. I’ve sent this import to identification and you should have matches ready soon.

[Mirjam]

Hi Anastasia,

Thank you very much for resolving the problem!

One more question: will the automated detector improve over time and with the images I upload, or has the training phase been completed and it will stay as it is now?

[Mirjam]

Hi Anastasia,

Thank you again! Most photos indeed had good automatic annotations. I tried to fix the rest.

Unfortunately, most of the bad automatic annotations cannot be deleted because they exceed the frame. I read in the forum that you are aware of this problem and are about to fix it.

Therefore, I deleted the photos with poor automatic annotations and uploaded them again as a new bulk import. This time, I skipped detection and identification to avoid a bad annotation being added to the photos. I then added an annotation manually, after which the identification process began automatically.

After a while, the identification process was finished for all photos, but the number of candidates against which the photos had been compared differed. For the photos that you edited and that had a good automatic annotation, the number of candidates was 363 after I started the identification process again via the bulk import webpage. For the photos to which I added manual annotations, the number of candidates ranged from 310 to 312. There were 4 photos that I added manual annotations to. The difference between 310 and 312 may be due to the fact that identification began immediately after the annotation was added. I added the annotations one by one, so while one annotation was complete, the other was still in progress. This meant that the first annotation might not have included the second annotation as a matching candidate.

The project to which I added the photos has 155 encounters (including all the new photos). I wonder why the number of candidates varies (363 on one hand and 312 on the other), given that all photos are part of the same project, and all identification jobs started with the same number of photos in the project and the same filters for identification (the project).

I would like to start the identification process again now that all photos have good annotations, because I want to be sure that all photos are included as matching candidates. However, it seems that this is impossible.

Since the latest Wildbook update, the ‘Start Match’ button on the project webpage unfortunately no longer works. Pressing it starts the matching process, but it only produces errors and never finishes (although it still works for photos uploaded before the update).

Also, I cannot start identification on the bulk import webpage for the photos where I skipped detection and identification during the bulk import. Even after manually adding an annotation to these photos, it still says that detection is not complete and therefore identification cannot begin.

Could you help me with this?

Thank you in advance!

[Anastasia]

Hi @Mirjam

@Mirjam

Unfortunately, most of the bad automatic annotations cannot be deleted because they exceed the frame. I read in the forum that you are aware of this problem and are about to fix it.

Because this is being flagged more frequently, I’m trying to see if the devs can move the fix up sooner as a patch.

@Mirjam

The difference between 310 and 312 may be due to the fact that identification began immediately after the annotation was added. I added the annotations one by one, so while one annotation was complete, the other was still in progress. This meant that the first annotation might not have included the second annotation as a matching candidate.

This is correct. If you post the link to the fixed import you want to send through identification again, I can push it through for you. Otherwise if it’s a small import, using the start a new match button will get you match results that include all of your current annotations.

@Mirjam

Since the latest Wildbook update, the ‘Start Match’ button on the project webpage unfortunately no longer works. Pressing it starts the matching process, but it only produces errors and never finishes (although it still works for photos uploaded before the update).

This sounds like a bug. Can you post the link to your project so I can look into it further?

@Mirjam

Also, I cannot start identification on the bulk import webpage for the photos where I skipped detection and identification during the bulk import. Even after manually adding an annotation to these photos, it still says that detection is not complete and therefore identification cannot begin.

This also sounds like a bug. I’ll need more time to research this and follow up on it.

[Mirjam]

Hi Anastasia,

Thank you!

The import that I would like to run through identification again only includes four photos, because I did not skip detection when uploading the other twenty that I uploaded in a different bulk import. Therefore, the ‘Send to identification’ button on the bulk import webpage works for those. I will therefore use the ‘Start a new match’ option for the four photos.

The link to one of my projects where the ‘Start match’ button on the project webpage no longer works is:

Amphibian and Reptile Wildbook

In this project the problem concerns all photos, even those uploaded before the update. Once the ‘Start Match’ button is pressed, the HotSpotter pattern-matcher produces errors and the MiewID v4 never finishes.

This is the link to the bulk import for which I skipped detection and identification when uploading it, and for which the ‘Send to identification’ button on the bulk import webpage does not work:

Amphibian and Reptile Wildbook

Thank you very much for your help!

[Anastasia]

Quick update: the issue with being unable to edit annotations has been fixed. You should now be able to edit or delete the annotations you previously couldn’t.

@Mirjam

This is the link to the bulk import for which I skipped detection and identification when uploading it, and for which the ‘Send to identification’ button on the bulk import webpage does not work:

Amphibian and Reptile Wildbook

Thanks! I forced these through identification and you should have matches ready to review now.

@Mirjam

The link to one of my projects where the ‘Start match’ button on the project webpage no longer works is:

Amphibian and Reptile Wildbook

In this project the problem concerns all photos, even those uploaded before the update. Once the ‘Start Match’ button is pressed, the HotSpotter pattern-matcher produces errors and the MiewID v4 never finishes.

In one example where I resent the match from the project page, I got Hotspotter matches back quickly, but I do see that MiewID never completes (here). Conversely, if I go directly to the encounter and restart the match from there, MiewID and Hotspotter results loaded (here).

I’ll write a ticket for this and post the link here when it’s ready. I know we planned to have a dedicated release for a redesign of the Projects workflow, so I suspect this would get bundled with those efforts. In the meantime as a workaround, try restarting your match from the encounter page instead of the project page.

[Anastasia]

ARW - MiewID results don't load when starting a match from the project page