I would have liked the video to be more in depth about facial recognition. How is the data processed in order to find those key feature of the human face? Does it uses trained neural network implemented in cameras and such?
***** Thanks for the reply! It pretty much validated how I thougth those worked. But I wasn't aware of the use of geometric model, or how they work in detail, I'll go and read on those. Thanks again, pal!
As a feedback: I can say that, a video which includes more smilings (like this one :) ) attracts more than the normal ones. Your videos are educating ones, so more smiling or jokes etc. make them easier and enjoyable to watch and learn.
That's nice. I am doing an undergrad summer research on Facial Expression Recognition and it is cool to see Computerphile talking about this topic. I would love to see more videos about i! :)
Nice overview ... it still would be interesting to get deeper into the actual algorithms though. If even modern crappy digital cameras all have pretty accurate facial detection on them, the algorithms must be relatively trivial these days. I'd love to see a video where they dig into how the algorithms actually go from that array of pixels to actually picking out the points and shapes.
I've really enjoyed these last couple Computerphiles, the vids are delving more into data processing, which is neat to see what the process has to deal in order for the output to work properly. Great stuff!
Could you do a video on how Googles Deep Dream works? Also, explain why it likes to see holes as eyes as well as other faces (dogs in most cases)? Is this related to visual pareidolia and why we see faces in objects and also eyes in the dark?
Interesting video! I have a question though. How much realism does it take for a facial recognition program to be able to recognize a portrait drawing or painting as a face? I remember uploading a digital drawing of mine to Facebook last year and it prompted me to tag the face. But my drawing style is kind of anime-ish with the big eyes and near non-existent nose, so I was wondering how it could recognize that.
ClaySmileSoil This entirely depends on the algorithm. A lot of it comes back to the distinction between geometric and appearance features, although it gets a bit more abstract. Basically, an algorithm that looks at relative positions of dark and light blobs (e.g. Viola & Jones) or edges can be 'fooled' into finding faces in drawings. However, systems that are trained with low-level appearance features (e.g. directly from pixel values) and are trained to only accept real faces will normally not detect drawings of faces. You can reinforce a system not to detect those by providing drawings of faces as negative examples.
TheNefari Some great work on that you can find here: www.cs.nott.ac.uk/~mfv/Documents/structpdJournal.pdf, and here: www.cs.nott.ac.uk/~yzt/pdf_files/tzimiro_CVPR15.pdf. It's dense material though, as it's far from easy. Maybe I can do one on this topic in the future?
Fabian Neundorf This may have been cut out. Erika Rosenbeg is a colleague who trained me in FACS, and she's the scientific advisor to Lie to Me. She's one of Paul Ekmans former PhD students.
Check out our channel, we really have some amazing footage, just read the descriptions in each of our videos to learn more about our company and the telecom industry. we also buy and sell new or used telecom equipment.
They need to work on a system that can detect sarcasm in RU-vid comments. It's a well-known fact that intelligence and ability to detect sarcasm are not correlated at all. Utterly useless.
I enjoy the channel, but don't like this kind of videos where the person talks of 95% of the time, I mean, with no animations or drawing on papers, or some kind of interaction!