An MXNet implementation of Fine-Grained Head Pose.
The original Pytorch version is here
We Do Not implement Dlib to detect face yet, input should be cropped head image with size of 48x48x1, we provide a pre-trained model here:
python code/test.py --mode 1You can also train on your datasets.
python train.pyOn a single TITAN X, it only takes about 5 minutes to finish 15 epochs on around 300k datasets. Also support for CPU mode.
The original method follows the paper.
We further exploits this problem and achieve state of the art performance. But due to secrecy agreement, the code will NOT be released until the project is finished or paper comes out. We only release the reproduced version and show some results right now.
{
author = {Haofan Wang}
institute = {Horizon Robotics inc}
contact = {[email protected]}
}