this is an interesting application - most likely you'd want it to let yourself in without a key. However, i'd argue that images from the porch of your house ca

this is an interesting application - most likely you'd want it to let yourself in without a key. However, i'd argue that images from the porch of your house can be very challenging - harsh illumination changes, pose, scale, facial deformations (emotion, make-up, etc), occlusions, and even weather (rain on the lense, fog, wind throwing your hair over your face, etc). You can really get a full gambit of difficulties we try to emulate in datasets - however there is a distinct advantage - the lack of imposters. You're not going to get more than a few hundred people entering through your front door every year (including yourself). This is actually very interesting. You can design an algorithm that fails with large data and still be very successful. I honestly cannot recall any papers that address this type of problem, particularly because every paper discusses the government and security uses. A method that robustly meets the needs of the average household, but is not under the same pressure to scale....interesting. I think reviewers would kill it though.