mxnet
|
channel pooling expression, do reduction over (local nearby) channels, used to implement local response normalization More...
#include <channel_unpool.h>
Public Member Functions | |
ChannelUnpoolingExp (const SrcExp &data_src, const SrcExp &data_pooled, const SrcExp &grad_pooled, index_t nsize, index_t kstride, index_t pad) | |
constructor More... | |
Public Member Functions inherited from mshadow::expr::MakeTensorExp< ChannelUnpoolingExp< Reducer, SrcExp, DType, srcdim >, SrcExp, srcdim, DType > | |
const ChannelUnpoolingExp< Reducer, SrcExp, DType, srcdim > & | real_self (void) const |
true self of subtype More... | |
Public Member Functions inherited from mshadow::expr::Exp< MakeTensorExp< ChannelUnpoolingExp< Reducer, SrcExp, DType, srcdim >, SrcExp, dim, DType >, DType, type::kChainer > | |
const MakeTensorExp< ChannelUnpoolingExp< Reducer, SrcExp, DType, srcdim >, SrcExp, dim, DType > & | self (void) const |
MakeTensorExp< ChannelUnpoolingExp< Reducer, SrcExp, DType, srcdim >, SrcExp, dim, DType > * | ptrself (void) |
Public Attributes | |
const SrcExp & | data_src_ |
source input, corresponds to src in pooling More... | |
const SrcExp & | data_pooled_ |
result of pooled data, corresponds to result of pooling More... | |
const SrcExp & | grad_pooled_ |
gradient data of pooled part, to be propgate down More... | |
index_t | pchannel_ |
channel of pooled expression More... | |
index_t | nsize_ |
kernel size in height More... | |
index_t | kstride_ |
kernel size in width More... | |
index_t | pad_ |
pad More... | |
Public Attributes inherited from mshadow::expr::MakeTensorExp< ChannelUnpoolingExp< Reducer, SrcExp, DType, srcdim >, SrcExp, srcdim, DType > | |
Shape< dim > | shape_ |
the shape of this expression More... | |
channel pooling expression, do reduction over (local nearby) channels, used to implement local response normalization
Reducer | reduction method during pooling |
SrcExp | source expression to be pooled from |
DType | the type of elements |
srcdim | dimension of src |
|
inline |
constructor
const SrcExp& mshadow::expr::ChannelUnpoolingExp< Reducer, SrcExp, DType, srcdim >::data_pooled_ |
result of pooled data, corresponds to result of pooling
const SrcExp& mshadow::expr::ChannelUnpoolingExp< Reducer, SrcExp, DType, srcdim >::data_src_ |
source input, corresponds to src in pooling
const SrcExp& mshadow::expr::ChannelUnpoolingExp< Reducer, SrcExp, DType, srcdim >::grad_pooled_ |
gradient data of pooled part, to be propgate down
index_t mshadow::expr::ChannelUnpoolingExp< Reducer, SrcExp, DType, srcdim >::kstride_ |
kernel size in width
index_t mshadow::expr::ChannelUnpoolingExp< Reducer, SrcExp, DType, srcdim >::nsize_ |
kernel size in height
index_t mshadow::expr::ChannelUnpoolingExp< Reducer, SrcExp, DType, srcdim >::pad_ |
pad
index_t mshadow::expr::ChannelUnpoolingExp< Reducer, SrcExp, DType, srcdim >::pchannel_ |
channel of pooled expression