Incubator-mxnet: How to understand terms "kNCHW", "NCW", "NCHW", "NCDHW", "kNHWC"...

Created on 11 Apr 2017  Â·  6Comments  Â·  Source: apache/incubator-mxnet

How to understand terms "kNCHW", "NCW", "NCHW", "NCDHW", "kNHWC", "kNCHW", "kNHWC".

Totally, lost.

Most helpful comment

The capital letter is used to specify the layout of data, which is used in convolution usually:
N: batch size
C: channel
H: height
W: width
so that "NCHW" means a data whose layout is (batch_size, channel, height, width)

and the lower case k represents a enum class, see: https://github.com/dmlc/mshadow/blob/master/mshadow/base.h#L312
https://github.com/dmlc/mxnet/blob/master/include/mxnet/operator.h#L24

All 6 comments

The capital letter is used to specify the layout of data, which is used in convolution usually:
N: batch size
C: channel
H: height
W: width
so that "NCHW" means a data whose layout is (batch_size, channel, height, width)

and the lower case k represents a enum class, see: https://github.com/dmlc/mshadow/blob/master/mshadow/base.h#L312
https://github.com/dmlc/mxnet/blob/master/include/mxnet/operator.h#L24

We need to document the behavior of layout.

Why "batch size" is N, and "k" represents "a enum calss" ?

@kernel8liang just custom, which is also common in other librarys

@sxjscience a pinned FAQ issue in github maybe also good

thanks.

Was this page helpful?
0 / 5 - 0 ratings