ONE - On-device Neural Engine
|
Describe how to build a depthwise convolution filter from a tensor. More...
#include <DepthwiseFilterCodec.h>
Public Member Functions | |
virtual | ~DepthwiseFilterEncoder ()=default |
virtual DepthwiseFilterShape | shape (const TensorShape &shape) const =0 |
virtual TensorIndex | value (const DepthwiseFilterIndex &index) const =0 |
Describe how to build a depthwise convolution filter from a tensor.
Let us assume that "enc" is a depthwise filter encoder.
Given a tensor "inp" and its shape "inp.shape", "enc" builds a depthwise filter "out" as follows:
for each valid filter_index for enc.shape(inp.shape) out.at(filter_index) = inp.at(enc.value(filter_index))
Definition at line 40 of file DepthwiseFilterCodec.h.
|
virtualdefault |
|
pure virtual |
|
pure virtual |
Implemented in loco::PermutingEncoder< Domain::DepthwiseFilter >.