AI SCIENCE
Even the experts can’t tell these leopards apart
Snow leopards aren’t called “ghosts of the mountains” for nothing.
Spotting one in the wild is so rare that even experts can spend months searching with no luck.
For years, researchers have tried to count how many are left in the wild, but the mix of rocky terrain, their natural camouflage, and unreliable tracking methods has made it nearly impossible.
Camera traps helped, automated devices that snap photos when there’s movement.
But sorting through thousands of images and trying to tell one leopard from another?
That’s where things got tricky. Each snow leopard has a unique coat pattern, but images can be blurry, taken from odd angles, or poorly lit, making matches slow and often inaccurate.
In fact, human error has led to the same leopard being counted more than once, inflating numbers by as much as 30%.
Here’s what you should know:
AI tools are speeding up the process of tracking snow leopards, but experts are still needed to confirm tricky matches.
Manual identification can be unreliable and inflate population numbers.
The most accurate results come from combining AI with human review, and it could be key to protecting these rare cats.
Tech’s got claws now
Now, AI is helping out.
Tools like HotSpotter and pose invariant embeddings can scan huge image libraries, recognise leopards by their coat patterns (even in awkward photos), and match them against a database.
When used together, they’ve hit an 85% accuracy rate.
These tools are now built into Whiskerbook.org, a free, open-source platform where researchers upload and review images.
That said, humans are still key. In one test, experts using Whiskerbook correctly identified 90% of leopards and estimated population size within just 3% of the real figure.
Without the tech, those results were much worse, especially for non-experts.
We gave robots facial recognition and they’re using it… to count cats??