Prediction: more clear structure of cruise model

modules/prediction/network/cruise_model/cruise_model.cc

@@ -15,14 +15,45 @@
  *****************************************************************************/
 
 #include "modules/prediction/network/cruise_model/cruise_model.h"
+#include "cyber/common/log.h"
 
 namespace apollo {
 namespace prediction {
 namespace network {
 
+using apollo::prediction::CruiseModelParameter;
+
 void CruiseModel::Run(const std::vector<Eigen::MatrixXf>& inputs,
                       Eigen::MatrixXf* output) const {
   // TODO(kechxu) implement
+  // Step 1: Run lane feature conv 1d
+
+  // Step 2: Run lane feature max pool 1d
+
+  // Step 3: Run lane feature avg pool 1d
+
+  // Step 4: Run obstacle feature fully connected
+
+  // Step 5: Concatenate [lane_feature, obstacle_feature]
+
+  // Step 6: Get classification result
+
+  // Step 7: Get regression result
+
+  // Step 8: Output
+}
+
+bool CruiseModel::LoadModel(
+    const CruiseModelParameter& cruise_model_parameter) {
+  CHECK(cruise_model_parameter.has_lane_feature_conv());
+  CHECK(cruise_model_parameter.has_lane_feature_maxpool());
+  CHECK(cruise_model_parameter.has_lane_feature_avgpool());
+  CHECK(cruise_model_parameter.has_obs_feature_fc());
+  CHECK(cruise_model_parameter.has_classify());
+  CHECK(cruise_model_parameter.has_regress());
+
+  cruise_model_parameter_.CopyFrom(cruise_model_parameter);
+  return true;
 }
 
 }  // namespace network

modules/prediction/network/cruise_model/cruise_model.h

@@ -30,6 +30,14 @@ namespace network {
 
 class CruiseModel : public NetModel {
  public:
+  /**
+   * @brief Load cruise network parameters from a protobuf message
+   * @param CruiseModelParameter is a protobuf message
+   * @return True if successfully loaded, otherwise False
+   */
+  bool LoadModel(
+      const apollo::prediction::CruiseModelParameter& cruise_model_parameter);
+
   /**
    * @brief Compute the model output from inputs according to a defined layers'
    * flow
@@ -38,6 +46,9 @@ class CruiseModel : public NetModel {
    */
   void Run(const std::vector<Eigen::MatrixXf>& inputs,
            Eigen::MatrixXf* output) const override;
+
+ private:
+  CruiseModelParameter cruise_model_parameter_;
 };
 
 }  // namespace network

modules/prediction/proto/cruise_model.proto

@@ -6,7 +6,7 @@ import "modules/prediction/proto/network_layers.proto";
 
 // Intermediate building blocks:
 
-message LaneFeatureConv {
+message LaneFeatureConvParameter {
   optional Conv1dParameter conv1d_0 = 1;
   optional ActivationParameter activation_1 = 2;
   optional Conv1dParameter conv1d_2 = 3;
@@ -14,14 +14,14 @@ message LaneFeatureConv {
   optional Conv1dParameter conv1d_4 = 5;
 }
 
-message ObsFeatureFC {
+message ObsFeatureFCParameter {
   optional DenseParameter linear_0 = 1;
   optional ActivationParameter activation_1 = 2;
   optional DenseParameter linear_3 = 3;
   optional ActivationParameter activation_4 = 4;
 }
 
-message Classify {
+message ClassifyParameter {
   optional DenseParameter linear_0 = 1;
   optional ActivationParameter activation_1 = 2;
   optional DenseParameter linear_3 = 3;
@@ -32,7 +32,7 @@ message Classify {
   optional ActivationParameter activation_10 = 8;
 }
 
-message Regress {
+message RegressParameter {
   optional DenseParameter linear_0 = 1;
   optional ActivationParameter activation_1 = 2;
   optional DenseParameter linear_3 = 3;
@@ -46,11 +46,11 @@ message Regress {
 
 // Final model
 
-message CruiseModel {
-  optional LaneFeatureConv lane_feature_conv = 1;
+message CruiseModelParameter {
+  optional LaneFeatureConvParameter lane_feature_conv = 1;
   optional MaxPool1dParameter lane_feature_maxpool = 2;
   optional AvgPool1dParameter lane_feature_avgpool = 3;
-  optional ObsFeatureFC obs_feature_fc = 5;
-  optional Classify classify = 6;
-  optional Regress regress = 7;
+  optional ObsFeatureFCParameter obs_feature_fc = 5;
+  optional ClassifyParameter classify = 6;
+  optional RegressParameter regress = 7;
 }