Perception: upgrade onnx models and adapt for newer offline training …

…program (ApolloAuto#11248)

modules/perception/lidar/lib/detection/lidar_point_pillars/nms_cuda.cu

@@ -80,7 +80,7 @@ __global__ void nms_kernel(const int n_boxes, const float nms_overlap_thresh,
     if (row_start == col_start) {
       start = threadIdx.x + 1;
     }
-    for (int i = start; i < col_size; i++) {
+    for (int i = start; i < col_size; ++i) {
       if (devIoU(cur_box, block_boxes + i * num_box_corners) >
           nms_overlap_thresh) {
         t |= 1ULL << i;
@@ -122,14 +122,14 @@ void NmsCuda::DoNmsCuda(const int host_filter_count,
   std::vector<uint64_t> remv(col_blocks);
   memset(&remv[0], 0, sizeof(uint64_t) * col_blocks);
 
-  for (int i = 0; i < host_filter_count; i++) {
+  for (int i = 0; i < host_filter_count; ++i) {
     int nblock = i / num_threads_;
     int inblock = i % num_threads_;
 
     if (!(remv[nblock] & (1ULL << inblock))) {
       out_keep_inds[(*out_num_to_keep)++] = i;
       uint64_t *p = &host_mask[0] + i * col_blocks;
-      for (int j = nblock; j < col_blocks; j++) {
+      for (int j = nblock; j < col_blocks; ++j) {
         remv[j] |= p[j];
       }
     }

modules/perception/lidar/lib/detection/lidar_point_pillars/point_pillars.h

@@ -104,6 +104,7 @@ class PointPillars {
   static const int kNumClass;
   static const int kMaxNumPillars;
   static const int kMaxNumPointsPerPillar;
+  static const int kNumPointFeature;
   static const int kPfeOutputSize;
   static const int kGridXSize;
   static const int kGridYSize;
@@ -174,22 +175,16 @@ class PointPillars {
   int* dev_cumsum_along_x_;
   int* dev_cumsum_along_y_;
 
-  float* dev_pillar_x_;
-  float* dev_pillar_y_;
-  float* dev_pillar_z_;
-  float* dev_pillar_i_;
-
-  float* dev_x_coors_for_sub_shaped_;
-  float* dev_y_coors_for_sub_shaped_;
-  float* dev_pillar_feature_mask_;
+  float* dev_pillar_point_feature_;
+  float* dev_pillar_coors_;
 
   float* dev_box_anchors_min_x_;
   float* dev_box_anchors_min_y_;
   float* dev_box_anchors_max_x_;
   float* dev_box_anchors_max_y_;
   int* dev_anchor_mask_;
 
-  void* pfe_buffers_[9];
+  void* pfe_buffers_[4];
   void* rpn_buffers_[4];
 
   float* dev_scattered_feature_;
@@ -274,15 +269,15 @@ class PointPillars {
 
   /**
    * @brief Preproces points
-   * @param[in] in_points_array pointcloud array
+   * @param[in] in_points_array Point cloud array
    * @param[in] in_num_points Number of points
    * @details Call CPU or GPU preprocess
    */
   void Preprocess(const float* in_points_array, const int in_num_points);
 
   /**
    * @brief Preproces by CPU
-   * @param[in] in_points_array pointcloud array
+   * @param[in] in_points_array Point cloud array
    * @param[in] in_num_points Number of points
    * @details The output from preprocessCPU is reproducible, while preprocessGPU
    * is not
@@ -291,34 +286,29 @@ class PointPillars {
 
   /**
    * @brief Preproces by GPU
-   * @param[in] in_points_array pointcloud array
+   * @param[in] in_points_array Point cloud array
    * @param[in] in_num_points Number of points
-   * @details Faster preprocess comapared with CPU preprocess
+   * @details Faster preprocess compared with CPU preprocess
    */
   void PreprocessGPU(const float* in_points_array, const int in_num_points);
 
   /**
    * @brief Convert anchors to box form like min_x, min_y, max_x, max_y anchors
-   * @param[in] anchors_px_ Represents x-coordinate value for a corresponding
-   * anchor
-   * @param[in] anchors_py_ Represents y-coordinate value for a corresponding
-   * anchor
-   * @param[in] anchors_dx_ Represents x-dimension value for a corresponding
-   * anchor
-   * @param[in] anchors_dy_ Represents y-dimension value for a corresponding
-   * anchor
-   * @param[in] box_anchors_min_x_ Represents minimum x value for a
-   * correspomding anchor
-   * @param[in] box_anchors_min_y_ Represents minimum y value for a
-   * correspomding anchor
-   * @param[in] box_anchors_max_x_ Represents maximum x value for a
-   * correspomding anchor
-   * @param[in] box_anchors_max_y_ Represents maximum y value for a
-   * correspomding anchor
+   * @param[in] anchors_px_
+   *   Represents x-coordinate value for a corresponding anchor
+   * @param[in] anchors_py_
+   *   Represents y-coordinate value for a corresponding anchor
+   * @param[in] box_anchors_min_x_
+   *   Represents minimum x value for a corresponding anchor
+   * @param[in] box_anchors_min_y_
+   *   Represents minimum y value for a corresponding anchor
+   * @param[in] box_anchors_max_x_
+   *   Represents maximum x value for a corresponding anchor
+   * @param[in] box_anchors_max_y_
+   *   Represents maximum y value for a corresponding anchor
    * @details Make box anchors for nms
    */
   void ConvertAnchors2BoxAnchors(float* anchors_px_, float* anchors_py_,
-                                 float* anchors_dx_, float* anchors_dy_,
                                  float* box_anchors_min_x_,
                                  float* box_anchors_min_y_,
                                  float* box_anchors_max_x_,
@@ -339,25 +329,23 @@ class PointPillars {
    * @param[in] nms_overlap_threshold IOU threshold for NMS
    * @param[in] pfe_onnx_file Pillar Feature Extractor ONNX file path
    * @param[in] rpn_onnx_file Region Proposal Network ONNX file path
-   * @details Variables could be chaned through rosparam
+   * @details Variables could be changed through point_pillars_detection
    */
-  PointPillars(const bool reproduce_result_mode,
-               const float score_threshold,
+  PointPillars(const bool reproduce_result_mode, const float score_threshold,
                const float nms_overlap_threshold,
                const std::string pfe_onnx_file,
                const std::string rpn_onnx_file);
   ~PointPillars();
 
   /**
    * @brief Call PointPillars for the inference
-   * @param[in] in_points_array Pointcloud array
+   * @param[in] in_points_array Point cloud array
    * @param[in] in_num_points Number of points
    * @param[out] out_detections Network output bounding box
    * @param[out] out_labels Network output object's label
    * @details This is an interface for the algorithm
    */
-  void DoInference(const float* in_points_array,
-                   const int in_num_points,
+  void DoInference(const float* in_points_array, const int in_num_points,
                    std::vector<float>* out_detections,
                    std::vector<int>* out_labels);
 };