Dlib: What does the 128D descriptor physically represent?
abhisheksoni27
All 4 comments
Nothing. It's just numbers and the more similar they are the more likely two images are of the same person.
davisking
on 6 Aug 2017
Can you take a moment and think about how you call them "just numbers" and then say they actually are related to a face? I am not trying to be pretentious. You obviously know more than me. (You created the awesome library! :tada: ) Maybe, you think the question is dumb. And I might as well hear that rather than an ambiguous answer.
abhisheksoni27
on 6 Aug 2017
There isn't any more to it than what I said. It's not anything like "the 16th element models the ears and the 42th element models the nose". It's a 128D vector where the Euclidean distance between them has this property of separating identities. That's the only structure built into the model.
davisking
on 6 Aug 2017
Thanks for the clarification, Sir.
abhisheksoni27
on 6 Aug 2017
Related issues
farazirfan47
路
5Comments
joeking11829
路
5Comments
AeroClassics
路
4Comments
alison-carrera-pegasus
路
3Comments
great-thoughts
路
5Comments
Most helpful comment
There isn't any more to it than what I said. It's not anything like "the 16th element models the ears and the 42th element models the nose". It's a 128D vector where the Euclidean distance between them has this property of separating identities. That's the only structure built into the model.