Open Dr4zin opened 11 months ago
It can be implemented, but age detection has very low accuracy especially when you try to detect children There will be a lot of incorrect detections, I already tested this feature and unfortunately current genderage model (as part of buffalo_l) is not very accurate in age detection
Not only age, but gender detection is also very unreliable.
It can be implemented, but age detection has very low accuracy especially when you try to detect children There will be a lot of incorrect detections, I already tested this feature and unfortunately current genderage model (as part of buffalo_l) is not very accurate in age detection
since dectectioon has low accuracy, then maybe it's even better to have option for manually specifying ? For the same person, I usually get 20 - 30 - 54 etc randomly. I might be worth just typing ourselves when it happens ?
Manually specifying does not make any sense - just think about it. Even gender detection is often wrong. Your only recourse is to specify the head using 0 or 1, or 2, etc. If two heads are even somewhat close to each other it fails totally and you get distorted pixels, and out of focus faces.
If we have the option to select gender, and that's currently not reliable, then why not remove it? Alternatively, give the option to 'unreliably' use an age/age range - it might work for some under certain conditions
Feature description
Since even the same subject often receives an age detection that differs a lot depending on the photo chosen, I wondered if it would be possible to specify the subject age (source face and target face) with an option, so that it can be "locked".