Jie Zhang

I don't recommend you to use CPU which will be very slow. What's more, you should change the `max_iter` in solver.prototxt.template under [prototxt](https://github.com/luoyetx/deep-landmark/tree/master/prototxt). The default `max_iter` is very large and...

@yyytq you may consider to modify `PYTHONPATH` environment. I don't know why you encounter this problem but I have no error on CentOS.

It gives a little better performance than `pixel / 256` or `(pixel - 128) / 128` when I train the network. There's no particular reason I choose this method. If...

pay attention to the training loss, level-1 is important. make sure your training loss is less than 1e-2 or 1e-3. Look at #3, #6 and #14, they all meet the...

@mzpmzk Maybe you need to carefully train the networks of level-3. Actually, I train the networks not by bootstrap.sh but using `caffe train` for every networks. You need to pay...

@pidanxhq pay attention to the training log, your networks for Right Eye doesn't not converge.

The code [here](https://github.com/luoyetx/deep-landmark/blob/master/webapp/landmark.py#L171) process the image. The model here is still not comparable to the original paper.

@XiaXuehai Yes. Caffe support an external mean_file for data layer. But I didn't use it. Actually, I did the image processing in my own code and give a raw data...

@hugowangyj You can find some data in this [pdf](https://github.com/luoyetx/face-alignment-presentation/blob/master/Face-Alignment.pdf).

clockwise rotation and anticlockwise rotation, 0.5 is the possibility to do the rotation.