bs-tensorRT/common.cpp at main · KellyGui/bs-tensorRT · GitHub

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
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
#include "common.h"

using namespace nvinfer1;

IScaleLayer* addBatchNorm2d(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, std::string lname, float eps) {
    float *gamma = (float*)weightMap[lname + ".weight"].values;
    float *beta = (float*)weightMap[lname + ".bias"].values;
    float *mean = (float*)weightMap[lname + ".running_mean"].values;
    float *var = (float*)weightMap[lname + ".running_var"].values;
    int len = weightMap[lname + ".running_var"].count;

    float *scval = reinterpret_cast<float*>(malloc(sizeof(float) * len));
    for (int i = 0; i < len; i++) {
        scval[i] = gamma[i] / sqrt(var[i] + eps);
    }
    Weights scale{ DataType::kFLOAT, scval, len };

    float *shval = reinterpret_cast<float*>(malloc(sizeof(float) * len));
    for (int i = 0; i < len; i++) {
        shval[i] = beta[i] - mean[i] * gamma[i] / sqrt(var[i] + eps);
    }
    Weights shift{ DataType::kFLOAT, shval, len };

    float *pval = reinterpret_cast<float*>(malloc(sizeof(float) * len));
    for (int i = 0; i < len; i++) {
        pval[i] = 1.0;
    }
    Weights power{ DataType::kFLOAT, pval, len };

    weightMap[lname + ".scale"] = scale;
    weightMap[lname + ".shift"] = shift;
    weightMap[lname + ".power"] = power;
    IScaleLayer* scale_1 = network->addScale(input, ScaleMode::kCHANNEL, shift, scale, power);
    assert(scale_1);
    return scale_1;
}

IActivationLayer* bottleneck(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, int inch, int outch, int stride, int dilation, std::string lname) {
    Weights emptywts{DataType::kFLOAT, nullptr, 0};

    IConvolutionLayer* conv1 = network->addConvolutionNd(input, outch, DimsHW{1, 1}, weightMap[lname + "conv1.weight"], emptywts);
    assert(conv1);

    IScaleLayer* bn1 = addBatchNorm2d(network, weightMap, *conv1->getOutput(0), lname + "bn1", 1e-5);

    IActivationLayer* relu1 = network->addActivation(*bn1->getOutput(0), ActivationType::kRELU);
    assert(relu1);

    IConvolutionLayer* conv2 = network->addConvolutionNd(*relu1->getOutput(0), outch, DimsHW{3, 3}, weightMap[lname + "conv2.weight"], emptywts);
    assert(conv2);
    conv2->setStrideNd(DimsHW{stride, stride});
    conv2->setPaddingNd(DimsHW{dilation, dilation});
    conv2->setDilation(DimsHW{dilation, dilation});

    IScaleLayer* bn2 = addBatchNorm2d(network, weightMap, *conv2->getOutput(0), lname + "bn2", 1e-5);

    IActivationLayer* relu2 = network->addActivation(*bn2->getOutput(0), ActivationType::kRELU);
    assert(relu2);

    IConvolutionLayer* conv3 = network->addConvolutionNd(*relu2->getOutput(0), outch * 4, DimsHW{1, 1}, weightMap[lname + "conv3.weight"], emptywts);
    assert(conv3);

    IScaleLayer* bn3 = addBatchNorm2d(network, weightMap, *conv3->getOutput(0), lname + "bn3", 1e-5);

    IElementWiseLayer* ew1;
    if (stride != 1 || inch != outch * 4) {
        IConvolutionLayer* conv4 = network->addConvolutionNd(input, outch * 4, DimsHW{1, 1}, weightMap[lname + "downsample.0.weight"], emptywts);
        assert(conv4);
        conv4->setStrideNd(DimsHW{stride, stride});

        IScaleLayer* bn4 = addBatchNorm2d(network, weightMap, *conv4->getOutput(0), lname + "downsample.1", 1e-5);
        ew1 = network->addElementWise(*bn4->getOutput(0), *bn3->getOutput(0), ElementWiseOperation::kSUM);
    } else {
        ew1 = network->addElementWise(input, *bn3->getOutput(0), ElementWiseOperation::kSUM);
    }
    IActivationLayer* relu3 = network->addActivation(*ew1->getOutput(0), ActivationType::kRELU);
    assert(relu3);
    return relu3;
}


IActivationLayer*  steamBlock(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, int outch, int ksize, int s, std::string lnameCon, std::string lnameBn){
    Weights emptywts{ DataType::kFLOAT, nullptr, 0 };
    int p = ksize / 3;
    IConvolutionLayer* conv1 = network->addConvolutionNd(input,outch,DimsHW{ksize,ksize}, weightMap[lnameCon], emptywts);
    assert(conv1);
    conv1->setStrideNd(DimsHW({s,s}));
    conv1->setPaddingNd(DimsHW({p,p}));

    IScaleLayer* bn1 = addBatchNorm2d(network,weightMap, *conv1->getOutput(0),lnameBn, 1e-5);
    assert(bn1);

    //ReLU
    auto relu = network->addActivation(*bn1->getOutput(0), ActivationType::kRELU);
    assert(relu);

    return relu;

}


IActivationLayer*  convBlock(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, int outch, int ksize, std::string lnameCon, std::string lnameBn, int s, int p, int d, int groups){
    Weights emptywts{ DataType::kFLOAT, nullptr, 0 };

    IConvolutionLayer* conv1 = network->addConvolutionNd(input,outch,DimsHW{ksize,ksize}, weightMap[lnameCon], emptywts);
    assert(conv1);
    conv1->setStrideNd(DimsHW({s,s}));
    conv1->setPaddingNd(DimsHW{p,p});
    conv1->setDilation(DimsHW{d,d});
    conv1->setNbGroups(groups);

    IScaleLayer* bn1 = addBatchNorm2d(network,weightMap, *conv1->getOutput(0),lnameBn, 1e-5);
    assert(bn1);

    //ReLU
    auto relu = network->addActivation(*bn1->getOutput(0), ActivationType::kRELU);
    assert(relu);

    return relu;

}

ILayer* resNetBlock(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input,  int outch, int ksize, int s, std::string lname, int expansion, ILayer *downSample){
    auto cv1 = convBlock(network, weightMap, input, outch, 1, lname+".conv1.weight", lname+".bn1"); //stride=1, padding=0
    auto cv2 = convBlock(network, weightMap, *cv1->getOutput(0), outch, 3, lname+".conv2.weight", lname+".bn2", s, 1);
    auto cv3 = convBlock(network, weightMap, *cv2->getOutput(0), outch*expansion, 1, lname+".conv3.weight", lname+".bn3");//stride =1, padding =0
    if(downSample!=nullptr){
        auto add = network->addElementWise(*cv3->getOutput(0),*downSample->getOutput(0), ElementWiseOperation::kSUM);
        auto relu = network->addActivation(*add->getOutput(0), ActivationType::kRELU);
        return relu;
    }
    auto add2 = network->addElementWise(*cv3->getOutput(0),input,ElementWiseOperation::kSUM);
    auto relu = network->addActivation(*add2->getOutput(0), ActivationType::kRELU);
    return relu;
}

ILayer* imagePool(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, int outch, std::string lname){
    auto avgPool = network->addPoolingNd(input, PoolingType::kAVERAGE, DimsHW{128,128});
    assert(avgPool);
    auto conv1 = convBlock(network, weightMap, *avgPool->getOutput(0), outch, 1, lname+".1.conv.weight", lname+".1.bn");
    assert(conv1);
    return conv1;
}

ILayer* sepBottleNeck(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, int outch, std::string lname){
    int inch = input.getDimensions().d[0];
    auto depthwiseConv1 = convBlock(network, weightMap, input, inch, 3, lname+".0.depthwise_conv.conv.weight", lname+".0.depthwise_conv.bn", 1, 1, 1, inch);
    auto pointwiseConv1 = convBlock(network, weightMap, *depthwiseConv1->getOutput(0), outch, 1, lname+".0.pointwise_conv.conv.weight", lname+".0.pointwise_conv.bn");
    inch = pointwiseConv1->getOutput(0)->getDimensions().d[0];
    auto depthwiseConv2 = convBlock(network, weightMap, *pointwiseConv1->getOutput(0), inch, 3, lname+".1.depthwise_conv.conv.weight", lname+".1.depthwise_conv.bn", 1, 1, 1, inch);
    auto pointwiseConv2 = convBlock(network, weightMap, *depthwiseConv2->getOutput(0), outch, 1, lname+".1.pointwise_conv.conv.weight", lname+".1.pointwise_conv.bn");
    return pointwiseConv2;
}