Suggestion: add the ability to search by photo on the site
In my idea, it could be done with a system similar to Google’s image search
You click the search button, and there will be something like a button for searching by photo; you upload a photo of a manul there, and some algorithm (I’m not versed in this topic) finds that photo on the site
I really hope this is possible, because I’m sure it will help a lot of people!!
Comments 13
On the one hand, yes...
On the other hand, when I feed Yandex a photo of a manul, it finds me all the manuls at once 😸
In theory, great! In practice, they’re internet photos that are supposed to be inserted into the site for searching, and most of the photos on the site will not match by 99.9999999999%.
On the site, for the most part, fresh photos from zoos uploaded by users are posted. And old, popular photos of manuls are circulating around the internet.
Анна, well, that’s true too, unfortunately
Timofey, Wait Timofey.. You can talk? Can any other manuls talk or are you the only one?
I really enjoy seeing other people posting and asking for identification. I got to learn about more manuls!
In short, theoretically, yes, it’s possible. What’s more, that was exactly what Roma talked about on the livestream on April 23, https://manulization.ru/manulover/vika/posts/results-of-the-giveaway-on-the-pallas-s-cat-livestream-crssfgrqryyiko4l.html
Exactly — toward the end there was a section about the scientific value of the Manulization archive and the possibility of training animal-recognition neural networks on this set of photo and video materials. The question is who will take on this work. And how soon, and what the result will be.
In short: yes, theoretically it is possible. No, not today and not tomorrow. This is work for several dissertations and, accordingly, several years. And, possibly, the main focus of the practical work will not be on manuls.
Pavel Burov, I mean not identifying a manul from a photo, but searching for photos on the site
Sugary Shampoo, well, there are no other manuls on this site, so I don’t know :)
Timofey, This task is suddenly almost more difficult. A photograph can be processed differently, rotated and/or cropped, mirrored, have text added, objects removed/retouched, something added/drawn in, turned into a collage, or simply recompressed into another format (possibly with conversion between color spaces with one or another colorimetric intent).
Roughly speaking, what is needed is a well-trained AI (for example, a neural network) that can decide whether photographs are sufficiently similar. Such neural nets exist now, but they are trained on large sets of photos of people, cities, and landscapes. It is not certain that they will work well on manuls.
But the task is interesting and, what is more important, practically useful. Sometimes people forget and upload the same photo to the site several times, and such a robot would make it possible to issue a warning: “The photo may be a duplicate of such-and-such photo already posted on the site.”
Pavel Burov, This is called a "bayanometer" 😁
Анна, [:|\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/|:]-o-meter
Yes, I like that idea, and it’s even in the plans. I mentioned it a little on stream. The basic idea is to train a neural network on a manulization dataset to distinguish individual manuls. Then you could make it so that on the site, when you upload a photo, the names would be filled in automatically right away; that’s first. And second, you could make a page like “what manul is this,” where you upload a photo and it tells you which manul it is. This neural network could also be used to find the same manuls in camera-trap photos.
One person suggested making something like that, but has disappeared for now.
I think it’s not super difficult, and part of this task could be vibe-coded, but it just takes time.
So, in short, this task is on the list :)
Roman Paulovverified_filled ,
Identifying manuls with a neural net is boring!
Fun is when everyone’s in it together! 😁
(though, I admit, when it comes to photos of manuls from camera traps, that’s a very useful feature).