how to improve the accuracy when detecting mood and age?
I am trying to do a project to analyze user's expression (i.e. Happy, sad, frustrated, etc. ) Currently only the detection of gender is accurate. For the mood and age it is not very accurate yet. Does anyone have any idea how to improve accuracy?
what are you using ? fisherfaces ? how large is your trainset ?
yea i am using fisherface. currently 20 for trained set... gender is working well, but for the detection of age and mood is really quite inaccurate! Any advice how to increase accuracy? :)
more train images then. say, 100 for each label.
gender is the most easy one here, just 2 labels.
so you mean i have to have bigger trained sets to improve accuracy for age and mood? so it means for happy/sad/angry etc. I have to ensure I have larger number of such images to improve accuracy? sorry i am new to this project and do not have much background information. Thanks for the help!
yes, exactly.
@SJ: Which kind of database is used for gender detection? Currently, I used AT database, but it is very bad result. Hence, I changed to other database.
AT is too small. try adience or faces94
Hi @berak , I tried adience but the X, Y, DX, DY is not working for me. Were you able to crop the faces using the data provided? Could you give me any help? Thank you.
@berak , here some samples I'm reading from the file 3 samples
hmm, it seems, the face detection already failed on those images