Damn, Amazon’s celebrity recogniser doesn’t think I’m famous
So why has Amazon built it?
Amazon Web Services (AWS) has just rolled out a new celebrity recognition tool, which developers can use to automatically identify famous people in images. I love tinkering with geeky Amazon tools, so naturally I had to give the demo version a crack.
The system first identifies a face in the picture, then using machine learning to try and identify if it's a famous person. It returns a name, a confidence score and an IMDb link if one is available. It had no trouble with one of finder's more frequently contemplated celebrities, The Block host Scott Cam:
This is good, since it suggests that the system isn't heavily US-centric (unlike some Amazon services we could mention). It also had no difficulty dealing with a picture featuring multiple 1980s pop stars, which I personally take as an excellent sign:
Unfortunately, fame has a threshold. While I pop up quite regularly on Australian TV screens, that clearly isn't enough for Amazon. When I threw a picture of myself into the system, the best it could suggest was that I was US journalist Andy Capper:
At least AWS was only 76% confident on that one. And I can draw some comfort from that fact the recogniser didn't do any better with finder.com.au's cofounder Fred Schebesta, who is much more famous than I am. Apparently he is actually Ryan Gosling:
My bruised ego aside, why has Amazon done this? As the blog explaining the launch notes: "If you have an image archive you can now index it by celebrity. You could also use a combination of celebrity recognition and object detection to build all kinds of search tools." I look forward to a tool which can automatically find every picture of famous people using iPhones.
Angus Kidman's Findings column looks at new developments and research that help you save money, make wise decisions and enjoy your life more. It appears Monday through Friday on finder.com.au.
- Health round-up: Bone health, healthcare satisfaction and IVF app
- Overwatch League attracts over 10 million views in its first week
- Finder To The Node: Only a quarter of FTTN NBN connections will reach 100Mbps
- Korea FTC Chair: Closing cryptocurency exchanges “unrealistic”
- Bitcoin and Ethereum are both underutilising their networks