Currently the input/output needs to be transposed.
This commit added the option to transpose weight matrix instead so can use original input/output directly.
It's been tested on SegNetBasic demo application.
But their is some performance drop, so I set the default to not use it.
Currently the input/output needs to be transposed. This commit added the option to transpose weight matrix instead so can use original input/output directly.
It's been tested on SegNetBasic demo application. But their is some performance drop, so I set the default to not use it.