Single_path_parse_resnet training method

[DanielXu123]

May I ask how to trained the single_path_resnet_celeba.caffemodel?

Is this trained without the GAN part and using the CelebA to train the Facial Attribute Prediction Net directly?

If so, could you pls also uppload the single path resnet training method, it will be very appreciated.

Meanwhile, to reduce the caffemodel size, how about replacing the Convolution layer to Depthwise Conv, will this cut the acc hardly?