inference-engine/src/inference_engine/builders/ie_detection_output_layer.cpp

   1 // Copyright (C) 2018-2019 Intel Corporation
   2 // SPDX-License-Identifier: Apache-2.0
   3 //
   4
   5 #include <builders/ie_detection_output_layer.hpp>
   6 #include <ie_cnn_layer_builder.h>
   7
   8 #include <cfloat>
   9 #include <vector>
  10 #include <string>
  11
  12 using namespace InferenceEngine;
  13
  14 Builder::DetectionOutputLayer::DetectionOutputLayer(const std::string& name): LayerDecorator("DetectionOutput", name) {
  15     getLayer()->getOutputPorts().resize(1);
  16     getLayer()->getInputPorts().resize(2);
  17     setBackgroudLabelId(-1);
  18 }
  19
  20 Builder::DetectionOutputLayer::DetectionOutputLayer(const Layer::Ptr& layer): LayerDecorator(layer) {
  21     checkType("DetectionOutput");
  22 }
  23
  24 Builder::DetectionOutputLayer::DetectionOutputLayer(const Layer::CPtr& layer): LayerDecorator(layer) {
  25     checkType("DetectionOutput");
  26 }
  27
  28 Builder::DetectionOutputLayer& Builder::DetectionOutputLayer::setName(const std::string& name) {
  29     getLayer()->setName(name);
  30     return *this;
  31 }
  32
  33 const std::vector<Port>& Builder::DetectionOutputLayer::getInputPorts() const {
  34     return getLayer()->getInputPorts();
  35 }
  36
  37 Builder::DetectionOutputLayer& Builder::DetectionOutputLayer::setInputPorts(const std::vector<Port> &ports) {
  38     if (ports.size() != 3)
  39         THROW_IE_EXCEPTION << "Incorrect number of inputs for DetectionOutput getLayer().";
  40     getLayer()->getInputPorts() = ports;
  41     return *this;
  42 }
  43
  44 const Port& Builder::DetectionOutputLayer::getOutputPort() const {
  45     return getLayer()->getOutputPorts()[0];
  46 }
  47
  48 Builder::DetectionOutputLayer& Builder::DetectionOutputLayer::setOutputPort(const Port &port) {
  49     getLayer()->getOutputPorts()[0] = port;
  50     return *this;
  51 }
  52
  53 size_t Builder::DetectionOutputLayer::getNumClasses() const {
  54     return getLayer()->getParameters().at("num_classes");
  55 }
  56 Builder::DetectionOutputLayer& Builder::DetectionOutputLayer::setNumClasses(size_t num) {
  57     getLayer()->getParameters()["num_classes"] = num;
  58     return *this;
  59 }
  60 int Builder::DetectionOutputLayer::getBackgroudLabelId() const {
  61     return getLayer()->getParameters().at("background_label_id");
  62 }
  63 Builder::DetectionOutputLayer& Builder::DetectionOutputLayer::setBackgroudLabelId(int labelId) {
  64     getLayer()->getParameters()["background_label_id"] = labelId;
  65     return *this;
  66 }
  67 int Builder::DetectionOutputLayer::getTopK() const {
  68     return getLayer()->getParameters().at("top_k");
  69 }
  70 Builder::DetectionOutputLayer& Builder::DetectionOutputLayer::setTopK(int topK) {
  71     getLayer()->getParameters()["top_k"] = topK;
  72     return *this;
  73 }
  74 int Builder::DetectionOutputLayer::getKeepTopK() const {
  75     return getLayer()->getParameters().at("keep_top_k");
  76 }
  77 Builder::DetectionOutputLayer& Builder::DetectionOutputLayer::setKeepTopK(int topK) {
  78     getLayer()->getParameters()["keep_top_k"] = topK;
  79     return *this;
  80 }
  81 int Builder::DetectionOutputLayer::getNumOrientClasses() const {
  82     return getLayer()->getParameters().at("num_orient_classes");
  83 }
  84 Builder::DetectionOutputLayer& Builder::DetectionOutputLayer::setNumOrientClasses(int numClasses) {
  85     getLayer()->getParameters()["num_orient_classes"] = numClasses;
  86     return *this;
  87 }
  88 std::string Builder::DetectionOutputLayer::getCodeType() const {
  89     return getLayer()->getParameters().at("code_type");
  90 }
  91 Builder::DetectionOutputLayer& Builder::DetectionOutputLayer::setCodeType(std::string type) {
  92     getLayer()->getParameters()["code_type"] = type;
  93     return *this;
  94 }
  95 int Builder::DetectionOutputLayer::getInterpolateOrientation() const {
  96     return getLayer()->getParameters().at("interpolate_orientation");
  97 }
  98 Builder::DetectionOutputLayer& Builder::DetectionOutputLayer::setInterpolateOrientation(int orient) {
  99     getLayer()->getParameters()["interpolate_orientation"] = orient;
 100     return *this;
 101 }
 102 float Builder::DetectionOutputLayer::getNMSThreshold() const {
 103     return getLayer()->getParameters().at("nms_threshold");
 104 }
 105 Builder::DetectionOutputLayer& Builder::DetectionOutputLayer::setNMSThreshold(float threshold) {
 106     getLayer()->getParameters()["nms_threshold"] = threshold;
 107     return *this;
 108 }
 109 float Builder::DetectionOutputLayer::getConfidenceThreshold() const {
 110     return getLayer()->getParameters().at("confidence_threshold");
 111 }
 112 Builder::DetectionOutputLayer& Builder::DetectionOutputLayer::setConfidenceThreshold(float threshold) {
 113     getLayer()->getParameters()["confidence_threshold"] = threshold;
 114     return *this;
 115 }
 116 bool Builder::DetectionOutputLayer::getShareLocation() const {
 117     return getLayer()->getParameters().at("share_location");
 118 }
 119 Builder::DetectionOutputLayer& Builder::DetectionOutputLayer::setShareLocation(bool flag) {
 120     getLayer()->getParameters()["share_location"] = flag;
 121     return *this;
 122 }
 123 bool Builder::DetectionOutputLayer::getVariantEncodedInTarget() const {
 124     return getLayer()->getParameters().at("variance_encoded_in_target");
 125 }
 126 Builder::DetectionOutputLayer& Builder::DetectionOutputLayer::setVariantEncodedInTarget(bool flag) {
 127     getLayer()->getParameters()["variance_encoded_in_target"] = flag;
 128     return *this;
 129 }
 130
 131 REG_VALIDATOR_FOR(DetectionOutput, [](const InferenceEngine::Builder::Layer::CPtr& input_layer, bool partial) {
 132     Builder::DetectionOutputLayer layer(input_layer);
 133     if (layer.getNumClasses() == 0) {
 134         THROW_IE_EXCEPTION << "NumClasses parameter is wrong in layer " << layer.getName() <<
 135                            ". It should be > 0.";
 136     }
 137     if (layer.getCodeType() != "caffe.PriorBoxParameter.CENTER_SIZE" &&
 138         layer.getCodeType() != "caffe.PriorBoxParameter.CORNER") {
 139         THROW_IE_EXCEPTION << "CodeType parameter is wrong in layer " << layer.getName() <<
 140                            ". It should be equal to 'caffe.PriorBoxParameter.CORNER' or 'caffe.PriorBoxParameter.CENTER_SIZE'";
 141     }
 142     if (layer.getBackgroudLabelId() < -1) {
 143         THROW_IE_EXCEPTION << "BackgroundLabelId parameter is wrong in layer " << layer.getName() <<
 144                            ". It should be >= 0 if this one is an Id of existing label else it should be equal to -1";
 145     }
 146     if (layer.getNMSThreshold() <= 0) {
 147         THROW_IE_EXCEPTION << "NMSThreshold parameter is wrong in layer " << layer.getName() <<
 148                            ". It should be > 0.";
 149     }
 150     if (layer.getConfidenceThreshold() <= 0) {
 151         THROW_IE_EXCEPTION << "ConfidenceThreshold parameter is wrong in layer " << layer.getName() <<
 152                            ". It should be > 0.";
 153     }
 154 });
 155
 156 REG_CONVERTER_FOR(DetectionOutput, [](const CNNLayerPtr& cnnLayer, Builder::Layer& layer) {
 157     layer.getParameters()["num_classes"] = static_cast<size_t>(cnnLayer->GetParamAsUInt("num_classes"));
 158     layer.getParameters()["background_label_id"] = cnnLayer->GetParamAsInt("background_label_id", 0);
 159     layer.getParameters()["top_k"] = cnnLayer->GetParamAsInt("top_k", -1);
 160     layer.getParameters()["keep_top_k"] = cnnLayer->GetParamAsInt("keep_top_k", -1);
 161     layer.getParameters()["num_orient_classes"] = cnnLayer->GetParamAsInt("num_orient_classes", 0);
 162     layer.getParameters()["code_type"] = cnnLayer->GetParamAsString("code_type", "caffe.PriorBoxParameter.CORNER");
 163     layer.getParameters()["interpolate_orientation"] = cnnLayer->GetParamAsInt("interpolate_orientation", 1);
 164     layer.getParameters()["nms_threshold"] = cnnLayer->GetParamAsFloat("nms_threshold");
 165     layer.getParameters()["confidence_threshold"] = cnnLayer->GetParamAsFloat("confidence_threshold", -FLT_MAX);
 166     layer.getParameters()["share_location"] = cnnLayer->GetParamsAsBool("share_location", true);
 167     layer.getParameters()["variance_encoded_in_target"] = cnnLayer->GetParamsAsBool("variance_encoded_in_target", false);
 168 });