IoT not matching my images

Hi there, I am working on the Internet of Turtles to match around 200 images, I am investigating the efficiency of the system for a research project. Within my uploads I know that there are 20 turtles which have been spotted across multiple years and they have been uploaded as separate observations, i know this due to having their flipper tag ID recorded therefore the IoT should be stating that a match has been found. I am working through my encounter uploads and selecting “match results” and on every encounter it states “Image Analysis has returned and no match was found.” despite knowing i have another submission of the same turtle. I am trying to determine why this may not be working, is this an error in the database or is it perhaps that i don’t have location ID set, i have the longitude/latitude and location description but as i have no location ID set is this why it is reporting no match found?

Hi @nesteban

Yes. When you don’t set a location ID and only provide lat/long coordinates, Wildbook will only look for match candidates of the same species with those specific coordinates. You can increase your candidate pool by choosing a Location ID from the drop-down menu when you edit the Encounter.

I had typed out a list of reasons this could be happening before I realized that I overlooked that you had no location ID listed. :sweat_smile: I’m going to leave this here anyway in case someone searching for help with a similar issue later finds it useful:

  • Location ID: Wildbooks search for match candidates within the same species and Location ID. So for example, if in one Encounter the Location ID is “Pacific Ocean” but in the other Encounter it’s “Pacific”, those are considered two unique locations and they won’t appear in each other’s matches.
  • Taxonomy: If the species listed between the two encounters don’t match, they won’t be in the same match candidate pool.
  • Viewpoint: IoT assigns a Viewpoint to each image based on which side of the animal the photo was taken. If one Encounter has images of a turtle’s left side and the other Encounter only has images of its right side, these images won’t appear in each other’s match results because IoT only compares similar viewpoints to each other.
  • Photo quality: If the image is taken in conditions where the turtle is obscured by low light or other objects, it’s harder for image analysis to make accurate guesses for match candidates.
  • Low match scores: Hotspotter only returns match results when the match score of the potential candidates are greater than 0. It won’t bother displaying results that rate lower than 0.

Okay thank you. If i work through my encounters and update the location ID’s will this rematch automatically in a few days or will i need to start a new match for each encounter. I have over 200 encounters so this would take a long time as when i tried before after changing the location ID it was loading for hours.
Additionally, I uploaded around 200 encounters but it is now stating i have over 400, i understand this happens when i create a new annotation but why have many of my encounters been split, for example one encounter is correct with all images and data present while another encounter has one picture from the correct one with no supporting data, if i deleted the repeat (incorrect one) where its been split would the other be removed alongside this?

If these we manually imported (via the Report an Encounter page), you would need to manually start a new match for each encounter. If they were part of a bulk import, you could re-send them to ID via the bulk import page without having to manually go into each Encounter.

Do you have a link to a specific Encounter I can look at? Typically, an Encounter is cloned when multiple annotations of the same type are detected (such as two head annotations or two body annotations) because Wildbook will assume that means there is more than one animal pictured and each Encounter represents one animal seen at a specific place and time. Sometimes detection makes a mistake and will create a cloned Encounter when there is only one animal in the photo. It’s ok to delete the cloned Encounter in this instance. It won’t affect your original encounter.

Thank you for your help,
I have approximately 200 uploads that I am matching, I have updated the location ID of each encounter and have started creating new matches so hopefully now they match correctly, however this match is taking ages to load each new match and then stops at “attempting to fetch results” but never eventually provides any results, as you can imagine this could take ages to work through all my encounters if it never loads. Is there any way this can be fixed or sped up? Here is an example of 2 encounters which should be the same individual but as different encounters as spotted at a different date. This is one that i have been waiting to load for a very long time.

Thank you

I see match results displayed for the first link you shared. For the second link, I restarted the match on the head annotation that was stuck at “attempting to fetch results” and it’s processing now. For reference, here’s our help doc on interpreting match result statuses: Matching FAQ - Wild Me Documentation

Depending on how many other jobs are in the queue, Hotspotter results can take longer to load than MiewID ones. Let me know if you continue to have trouble viewing results.

Thank you, the first link worked but when i “start new match” for all of of my other encounters it is still stuck not loading or attempting to fetch results.

How long after you’ve started a new match do you consider it stuck? Depending on how many other jobs are in the queue, Hotspotter can take anywhere from a few minutes to a few hours to load results. MiewID results will often load faster even when there are a lot of waiting jobs.

I haven’t seen it finish loading yet and it’s been days now, this is only for one encounter and I have 195 to match. I need to look at both MiewID and Hotspotter results. I understand it could take a few hours to load but I have a lot to match and they are taking considerable amounts of time.

Ok, thanks for clarifying. Do you have links to some of the other encounters you’re having trouble with that I can take a look at?

Thank you, I appreciate this being looked into, these are a few examples of encounters I have tried to start a new match for however none of my uploads under my account are loading and I have many to sort through for a research study, it is not just these few that aren’t loading.

1 Like

No problem; thanks for the links! I need to spend more time researching this. I hope to check in tomorrow with any updates.

Hi there. Checking in to let you know I don’t have any updates yet. It appears something is misbehaving in WBIA but am waiting for one of my teammates to be back in the office tomorrow to dig in more.

Hi @nesteban

We’ve figured out the cause of the issue, but it’s going to take us a little longer to fix. I’m sorry about the continued delay. Thanks for your patience.

Thank you for looking into this, do you have any idea roughly how long this may take?

We should have a resolution this week. Several members of our machine learning team are working on it. I’ll let you know as soon as I have any news.

Checking in to say we’re still working on resolving this. Thanks for your patience.

Quick update: We had to disable Hotspotter as a default algorithm for Hawksbill and Green sea turtles. It was contributing to long queues and taking up a lot of disk space, which in turn can cause jobs to fail.

Hotspotter can still be selected if you re-run matches manually, but removing it as a default algorithm should help things move faster in IoT overall. Many IoT jobs are successfully completing.

Thanks for your patience. I’ve verified that match results are displaying now for your example encounters.