[methehack]

It's not really a competition. Yet. Anyway, I think you're missing the point. There's a great mystery here. How can the human mind do this? "Facial recognition software managed to identify one suspect of the 4,000 captured by security cameras during the London riots. Constable Collins identified 180". And dollars to donuts Constable Collins used a fraction of the energy too -- maybe even a just a donut's worth -- he is a copper after all.

[lips]

Indeed. My first thought was "Are tech companies trying to poach these folks in order to try and deconstruct what cues they're using, and/or to create a corpus of verified matches from low-quality source?"

[fauigerzigerk]

> How can the human mind do this?

I think the most interesting point about this is that the human mind doesn't know how the human mind does it.

I wonder what kind of software they compared to Constable Collins. Did that software use designed features or learned features? What sort of data set was it trained on?

It wouldn't surprise me if current state of the art software that learns features from real CCTV footage fared a lot better than what they were using in 2011.

[tedks]

I'm sure his false positive rate is zero, and that he's never caused an innocent person to be arrested or imprisoned. After all, he's in the top 1% on a test designed by people from Harvard. Yes, Harvard. Thank the lord that once a policeman's fingered a known criminal, a jury will never disagree with them. I bet his conviction rate is stellar.

[PepeGomez]

I doubt they arrest somebody just based on that.

[tedks]

They named a person in the article based just on this police officer's word. He might never be able to get a job again if this article stays on the first page of google hits for his name. Who knows if he's the same person? Is there anything else it could be based on?