What Wildbook should this feature be in?
Flukebook
What would you like to see?
Hi all, it’s more a question, than a request for now. I’m not yet on fluke book, and I’m new to the community. I just used the docker version so far to see what it’s like, and I must say I like it and would like to go online, just needing my boss approval before making the jump.
First, thanks for all the great work, you guys open so many possibilities!
Which brings me to it, are blue whales supported on Flukebook? If not could it be possible to identify blue whales with hot spotter or Modified Groth and I3S?
I work on Humpback, Fin and Blue whales so I have questions regarding all of these species.
For fin whales, I have seen you started supporting matching with FinFindr, do you plan on implanting coloration pattern analysis also (chevron) as sometimes dorsal fins just have no marks but they can all be identified with the right coloration pattern?
Last but very important, is it possible or would it be possible to match humpback whales aerial (drone) pictures of the dorsal side of the fluke thanks to the trailing edge to traditional boat based pictures? This apply to the other 2 species also, for the aerial and horizontal match part.
How would this functionality help you?
I am about to start a project to see if photo-identification of these 3 species with aerial pictures is more effective than traditional boat based photo-ID as it would reduce the disturbance on the animals in the end and produce datas of a better quality I think. Hence all the questions. This way we could match easily the two kind of catalogues and also be able to keep track of the ID obtained by photos of citizens contributors.
Which brings me to it, are blue whales supported on Flukebook? If not could it be possible to identify blue whales with hot spotter or Modified Groth and I3S?
They are not, but we know how to tackle that challenge if you can gather the data. We would apply Pose Invariant Embeddings (PIE v2), which applies ML to learn how to extract individual features per species. PIE has been successfully applied by us to gray whales, orcas, whale sharks, right whales, and more.
To add blue whales, we would need 2000 photos (random viewpoints, dates, and lighting conditions) to train the ML detector. Then for individual ID (assuming we use PIE, OOTB HotSpotter will provide some base matching capability too), 3-5+ images per flank per animal for a 200+ individual blue whales.
do you plan on implanting coloration pattern analysis also (chevron) as sometimes dorsal fins just have no marks but they can all be identified with the right coloration pattern?
We have not run this analysis, but this too might be a good challenge for PIE v2. I’ll review the fin whale data in Flukebook and see if this might be a possibility.
Last but very important, is it possible or would it be possible to match humpback whales aerial (drone) pictures of the dorsal side of the fluke thanks to the trailing edge to traditional boat based pictures?
Would you be able to demonstrate how these images may differ from the boat-based?
Hi Jason,
Thank you very much for the quick and detailed response.
As I don’T have any footage of my own I extracted pictures from a youtube video from @DolphinDroneDom, I don’t own the copyright of these images I added the copyright to the pictures, if it is not allowed to do that please remove my message as I don’t want to do any harm, though I think it is okay as it just to illustrate a purpose and not to be used in any lucrative way.
The first picture kind of show you what you would get boat based (the whale was rolling and the drone could film the ventral side of the fluke). The second picture shows you exactly what we would get with a drone every time a Humpback dive. So I think you kind of need to flip the picture to have the trailing edge to be on the same side, or switch the order of the data points found by Curvrank…this is beyond my capacities, you know better than me. Do you think it is possible and could be done without too much work?
Of course we would try to be positioned perfectly above the whale so the fluke is in the perfect position (less angle possible)
I have checked our catalogue, cause I thought we might have been able to offer some help regarding that matter, but I’m actually very far from the numbers you need, most of our individuals and pictures are from the early 90’s! Meaning black and white pictures with low quality compared to what we see today. Sorry about that.
Just made a request to join Flukebook, so I should be adding Humpback pictures pretty soon.
