permute
permute 干活: 把 卷积层的输出N/C/H/W 弄成 T/N/D , 输入给LSTM
(N=number of samples, C=number of channels, usually 1 or 3, H=height of images, W=width of images)
N: number of samples
C: number of channels
H: height
W: width
(time_steps, n_samples, dim_input)
T: time_steps
N: n_samples
D: dim_input
https://www.zhihu.com/question/41949741/answer/124898736
https://groups.google.com/forum/#!topic/caffe-users/9WHDx0Avrzo