opencog / TinyCog

Small Robot, Toy Robot platform
33 stars 28 forks source link

facial expression #9

Closed Geta-meko closed 6 years ago

smigad commented 6 years ago

what's the point of being able to tell the naturalness of a person's face? We just need ratio facial features so that we don't have to set a static threshold for classifying a person's mouth or eyes are open/closed , wide/narrow... for example to tell when an eye is closed, one can just look at the difference between the upper and lower landmark points of the eye but that wouldn't work for two people with different eye sizes and that's where the need to also consider the width of the eye comes from... so that it is the compared to the width before saying it's open or not. are you planning on doing something like that?

Geta-meko commented 6 years ago

my idea is that once the camera sees a face it works the ratio of that particular face and compares it with the one we gave him as natural face. For example if the camera is given a smiley face it takes the ratio of the upper and lower landmarks of the lip and verses the upper landmark of lip to the bottom of chin. if the ratio is somewhat grater than to that of the natural face it will understand the face is smiley. we can implement it using switch... this was my plan

On 7 June 2018 at 11:07, Dagim Sisay notifications@github.com wrote:

what's the point of being able to tell the naturalness of a person's face? We just need ratio facial features so that we don't have to set a static threshold for classifying a person's mouth or eyes are open/closed , wide/narrow... for example to tell when an eye is closed, one can just look at the difference between the upper and lower landmark points of the eye but that wouldn't work for two people with different eye sizes and that's where the need to also consider the width of the eye comes from... so that it is the compared to the width before saying it's open or not. are you planning on doing something like that?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/opencog/TinyCog/pull/9#issuecomment-395331343, or mute the thread https://github.com/notifications/unsubscribe-auth/Aj6XGpmrYZRHdqNAdjiD2g209-IqEAPFks5t6N8-gaJpZM4Ud5-E .

smigad commented 6 years ago

great!

Geta-meko commented 6 years ago

by the way i used some landmarks from the 68 lms model. but i will edit it as soon as u included those i suggested last week

smigad commented 6 years ago

I couldn't train a new model this week... will send you a 42 points one today. Training will be done in a couple of hours.