Introduction
A project that applies CNN to simultaneously predict facial attributes, including gender and age. This can be extended to more attributes.
This multitask CNN has been trained on scratch from Celebrity dataset released by CUHK. After 500000 iterations, it reaches 84.21% for age classification and 95.47% for gender classification.
Evaluation Results
Male: M; Female: F; Young: Y; Old: O
Along with the attribute symbols, the network also outputs the probability associated with each attribute.
Usage
Install caffe and dlib, run the following command
python predict_vanilla_fd_attribute.py ../train_multitask/model/_iter_500000.caffemodel family3.jpg