I have 10K images, all unlabelled and raw, my task is to find the wrinkle score present on images.
Any hint how should I proceed, precision of algorithm I created is not very good, sometime it is working sometime it is not.
right now meauring the wrinkle is more like counting the number of edges on face.
Should I finally start thinking for neural network!
could you provide some sample images
just train a neural network to replicate the scoring that a human would give, then have humans label a couple/dozen/hundred pictures and see if that’s good enough.
Sure the images look like this
how about I label them binary one, and present the softmax output as percentage of wrinkles
@sankalp did you get any progress ?
Yes, opencv and dlib helped me in building the architecture, right now it is more statistic oriented, I will use deep learning arch to capture the wide variance of facial data, but that will not improve the recall without supervision learning.
For now job is getting done.