-
Notifications
You must be signed in to change notification settings - Fork 0
/
MWConvLayerImpl.hpp
60 lines (50 loc) · 2 KB
/
MWConvLayerImpl.hpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
/* Copyright 2018 The MathWorks, Inc. */
#ifndef _MW_CONV_LAYER_IMPL
#define _MW_CONV_LAYER_IMPL
#include "MWCNNLayerImpl.hpp"
// Convolution2DWCNNLayer
class MWConvLayerImpl : public MWCNNLayerImpl {
private:
int GFienSVKLlDQuZeqAdLC;
int GnxRkpzrPZimKtYYHSuG;
int IwKnaBoXVubIRYcxEJLH;
NEConvolutionLayer m_convLayer; // used for Convolution/1st half of grouped conv
NEConvolutionLayer m_convLayerSecondGroup; // used for 2nd half of grouped conv
Tensor m_convLayerWgtTensor;
Tensor m_convLayerBiasTensor;
SubTensor* m_prevLayer1; // subtensor for current layer input (1st half in grp conv)
SubTensor* m_prevLayer2; // subtensor for current layer input (2nd half in grp conv)
SubTensor* m_curLayer1; // subtensor for current layer output (1st half in grp conv)
SubTensor* m_curLayer2; // subtensor for current layer output (2nd half in grp conv)
SubTensor* m_convLayerWgtMWTensor; // subtensor for conv weights (1st half in grp conv)
SubTensor* m_convLayerWgtTensor2; // subtensor for conv weights (2nd half in grp conv)
SubTensor* m_convLayerBiasMWTensor; // subtensor for conv bias (1st half in grp conv)
SubTensor* m_convLayerBiasTensor2; // subtensor for conv bias (2nd half in grp conv)
void createConvLayer(int, int, int, int, int, int, int, int, const char*, const char*, int);
void allocate();
void predict();
void cleanup();
void loadWeights(const char*);
void loadBias(const char*);
public:
MWConvLayerImpl(MWCNNLayer*,
MWTargetNetworkImpl*,
int,
int,
int,
int,
int,
int,
int,
int,
int,
int,
int,
int,
int,
const char*,
const char*,
int );
~MWConvLayerImpl();
};
#endif