!5744 add DetectionPostProcess op

Merge pull request !5744 from wangzhe/master

mindspore/lite/nnacl/detection_post_process_parameter.h

@@ -0,0 +1,43 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef MINDSPORE_LITE_NNACL_DETECTION_POST_PROCESS_PARAMETER_H_
+#define MINDSPORE_LITE_NNACL_DETECTION_POST_PROCESS_PARAMETER_H_
+#include "nnacl/op_base.h"
+
+typedef struct DetectionPostProcessParameter {
+  OpParameter op_parameter_;
+  float h_scale_;
+  float w_scale_;
+  float x_scale_;
+  float y_scale_;
+  float nms_iou_threshold_;
+  float nms_score_threshold_;
+  int64_t max_detections_;
+  int64_t detections_per_class_;
+  int64_t max_classes_per_detection_;
+  int64_t num_classes_;
+  bool use_regular_nms_;
+  bool out_quantized_;
+
+  void *decoded_boxes_;
+  void *nms_candidate_;
+  void *selected_;
+  void *score_with_class_;
+  void *score_with_class_all_;
+} DetectionPostProcessParameter;
+
+#endif  // MINDSPORE_LITE_NNACL_DETECTION_POST_PROCESS_PARAMETER_H_

mindspore/lite/nnacl/fp32/detection_post_process.c

@@ -0,0 +1,220 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "nnacl/fp32/detection_post_process.h"
+#include <math.h>
+#include "nnacl/errorcode.h"
+#include "nnacl/op_base.h"
+
+int ScoreWithIndexCmp(const void *a, const void *b) {
+  ScoreWithIndex *pa = (ScoreWithIndex *)a;
+  ScoreWithIndex *pb = (ScoreWithIndex *)b;
+  if (pa->score > pb->score) {
+    return -1;
+  } else if (pa->score < pb->score) {
+    return 1;
+  } else {
+    return 0;
+  }
+}
+
+float IntersectionOverUnion(const BboxCorner *a, const BboxCorner *b) {
+  const float area_a = (a->ymax - a->ymin) * (a->xmax - a->xmin);
+  const float area_b = (b->ymax - b->ymin) * (b->xmax - b->xmin);
+  if (area_a <= 0 || area_b <= 0) {
+    return 0.0f;
+  }
+  const float ymin = a->ymin > b->ymin ? a->ymin : b->ymin;
+  const float xmin = a->xmin > b->xmin ? a->xmin : b->xmin;
+  const float ymax = a->ymax < b->ymax ? a->ymax : b->ymax;
+  const float xmax = a->xmax < b->xmax ? a->xmax : b->xmax;
+  const float h = ymax - ymin > 0.0f ? ymax - ymin : 0.0f;
+  const float w = xmax - xmin > 0.0f ? xmax - xmin : 0.0f;
+  const float inter = h * w;
+  return inter / (area_a + area_b - inter + 1e-8);
+}
+
+void DecodeBoxes(const int num_boxes, const float *input_boxes, const float *anchors, const BboxCenter scaler,
+                 float *decoded_boxes) {
+  for (int i = 0; i < num_boxes; ++i) {
+    BboxCenter *box = (BboxCenter *)(input_boxes + i * 4);
+    BboxCenter *anchor = (BboxCenter *)(anchors + i * 4);
+    BboxCorner *decoded_box = (BboxCorner *)(decoded_boxes + i * 4);
+    float y_center = box->y / scaler.y * anchor->h + anchor->y;
+    float x_center = box->x / scaler.x * anchor->w + anchor->x;
+    float h_half = 0.5f * expf(box->h / scaler.h) * anchor->h;
+    float w_half = 0.5f * expf(box->w / scaler.w) * anchor->w;
+    decoded_box->ymin = y_center - h_half;
+    decoded_box->xmin = x_center - w_half;
+    decoded_box->ymax = y_center + h_half;
+    decoded_box->xmax = x_center + w_half;
+  }
+}
+
+int NmsSingleClass(const int candidate_num, const float *decoded_boxes, const int max_detections,
+                   ScoreWithIndex *score_with_index, int *selected, const DetectionPostProcessParameter *param) {
+  uint8_t *nms_candidate = param->nms_candidate_;
+  const int output_num = candidate_num < max_detections ? candidate_num : max_detections;
+  int possible_candidate_num = candidate_num;
+  int selected_num = 0;
+  qsort(score_with_index, candidate_num, sizeof(ScoreWithIndex), ScoreWithIndexCmp);
+  for (int i = 0; i < candidate_num; ++i) {
+    nms_candidate[i] = 1;
+  }
+  for (int i = 0; i < candidate_num; ++i) {
+    if (possible_candidate_num == 0 || selected_num >= output_num) {
+      break;
+    }
+    if (nms_candidate[i] == 0) {
+      continue;
+    }
+    selected[selected_num++] = score_with_index[i].index;
+    nms_candidate[i] = 0;
+    possible_candidate_num--;
+    for (int t = i + 1; t < candidate_num; ++t) {
+      if (nms_candidate[t] == 1) {
+        const BboxCorner *bbox_i = (BboxCorner *)(decoded_boxes) + score_with_index[i].index;
+        const BboxCorner *bbox_t = (BboxCorner *)(decoded_boxes) + score_with_index[t].index;
+        const float iou = IntersectionOverUnion(bbox_i, bbox_t);
+        if (iou > param->nms_iou_threshold_) {
+          nms_candidate[t] = 0;
+          possible_candidate_num--;
+        }
+      }
+    }
+  }
+  return selected_num;
+}
+
+int NmsMultiClassesRegular(const int num_boxes, const int num_classes_with_bg, const float *decoded_boxes,
+                           const float *input_scores, float *output_boxes, float *output_classes, float *output_scores,
+                           const DetectionPostProcessParameter *param) {
+  const int first_class_index = num_classes_with_bg - (int)(param->num_classes_);
+  int *selected = (int *)(param->selected_);
+  ScoreWithIndex *score_with_index_single = (ScoreWithIndex *)(param->score_with_class_);
+  int all_classes_sorted_num = 0;
+  int all_classes_output_num = 0;
+  ScoreWithIndex *score_with_index_all = (ScoreWithIndex *)(param->score_with_class_all_);
+  for (int j = first_class_index; j < num_classes_with_bg; ++j) {
+    int candidate_num = 0;
+    // process single class
+    for (int i = 0; i < num_boxes; ++i) {
+      const float score = input_scores[i * num_classes_with_bg + j];
+      if (score >= param->nms_score_threshold_) {
+        score_with_index_single[candidate_num].score = score;
+        score_with_index_single[candidate_num++].index = i;
+      }
+    }
+    int selected_num = NmsSingleClass(candidate_num, decoded_boxes, param->detections_per_class_,
+                                      score_with_index_single, selected, param);
+    // process all classes
+    for (int i = 0; i < selected_num; ++i) {
+      // store class to index
+      score_with_index_all[all_classes_sorted_num].index = selected[i] * num_classes_with_bg + j;
+      score_with_index_all[all_classes_sorted_num++].score = input_scores[selected[i] * num_classes_with_bg + j];
+    }
+    all_classes_output_num =
+      all_classes_sorted_num < param->max_detections_ ? all_classes_sorted_num : param->max_detections_;
+    qsort(score_with_index_all, all_classes_sorted_num, sizeof(ScoreWithIndex), ScoreWithIndexCmp);
+    all_classes_sorted_num = all_classes_output_num;
+  }
+  for (int i = 0; i < param->max_detections_ * param->max_classes_per_detection_; ++i) {
+    if (i < all_classes_output_num) {
+      const int box_index = score_with_index_all[i].index / num_classes_with_bg;
+      const int class_index = score_with_index_all[i].index - box_index * num_classes_with_bg - first_class_index;
+      *((BboxCorner *)(output_boxes) + i) = *((BboxCorner *)(decoded_boxes) + box_index);
+      output_classes[i] = (float)class_index;
+      output_scores[i] = score_with_index_all[i].score;;
+    } else {
+      ((BboxCorner *)(output_boxes) + i)->ymin = 0;
+      ((BboxCorner *)(output_boxes) + i)->xmin = 0;
+      ((BboxCorner *)(output_boxes) + i)->ymax = 0;
+      ((BboxCorner *)(output_boxes) + i)->xmax = 0;
+      output_classes[i] = 0.0f;
+      output_scores[i] = 0.0f;
+    }
+  }
+  return all_classes_output_num;
+}
+
+int NmsMultiClassesFast(const int num_boxes, const int num_classes_with_bg, const float *decoded_boxes,
+                        const float *input_scores, float *output_boxes, float *output_classes, float *output_scores,
+                        const DetectionPostProcessParameter *param) {
+  const int first_class_index = num_classes_with_bg - (int)(param->num_classes_);
+  const int64_t max_classes_per_anchor =
+    param->max_classes_per_detection_ < param->num_classes_ ? param->max_classes_per_detection_ : param->num_classes_;
+  int candidate_num = 0;
+  ScoreWithIndex *score_with_class_all = (ScoreWithIndex *)(param->score_with_class_all_);
+  ScoreWithIndex *score_with_class = (ScoreWithIndex *)(param->score_with_class_);
+  int *selected = (int *)(param->selected_);
+  int selected_num;
+  int output_num = 0;
+  for (int i = 0; i < num_boxes; ++i) {
+    for (int j = first_class_index; j < num_classes_with_bg; ++j) {
+      float score_t = *(input_scores + i * num_classes_with_bg + j);
+      score_with_class_all[i * param->num_classes_ + j - first_class_index].score = score_t;
+      // save box and class info to index
+      score_with_class_all[i * param->num_classes_ + j - first_class_index].index = i * num_classes_with_bg + j;
+    }
+    qsort(score_with_class_all + i * param->num_classes_, param->num_classes_, sizeof(ScoreWithIndex),
+          ScoreWithIndexCmp);
+    const float score_max = (score_with_class_all + i * param->num_classes_)->score;
+    if (score_max >= param->nms_score_threshold_) {
+      score_with_class[candidate_num].index = i;
+      score_with_class[candidate_num++].score = score_max;
+    }
+  }
+  selected_num =
+    NmsSingleClass(candidate_num, decoded_boxes, param->max_detections_, score_with_class, selected, param);
+  for (int i = 0; i < selected_num; ++i) {
+    const ScoreWithIndex *box_score_with_class = score_with_class_all + selected[i] * param->num_classes_;
+    const int box_index = box_score_with_class->index / num_classes_with_bg;
+    for (int j = 0; j < max_classes_per_anchor; ++j) {
+      *((BboxCorner *)(output_boxes) + output_num) = *((BboxCorner *)(decoded_boxes) + box_index);
+      output_scores[output_num] = (box_score_with_class + j)->score;
+      output_classes[output_num++] =
+        (float)((box_score_with_class + j)->index % num_classes_with_bg - first_class_index);
+    }
+  }
+  for (int i = output_num; i < param->max_detections_ * param->max_classes_per_detection_; ++i) {
+    ((BboxCorner *)(output_boxes) + i)->ymin = 0;
+    ((BboxCorner *)(output_boxes) + i)->xmin = 0;
+    ((BboxCorner *)(output_boxes) + i)->ymax = 0;
+    ((BboxCorner *)(output_boxes) + i)->xmax = 0;
+    output_scores[i] = 0;
+    output_classes[i] = 0;
+  }
+  return output_num;
+}
+
+int DetectionPostProcess(const int num_boxes, const int num_classes_with_bg, float *input_boxes, float *input_scores,
+                         float *input_anchors, float *output_boxes, float *output_classes, float *output_scores,
+                         float *output_num, DetectionPostProcessParameter *param) {
+  BboxCenter scaler;
+  scaler.y = param->y_scale_;
+  scaler.x = param->x_scale_;
+  scaler.h = param->h_scale_;
+  scaler.w = param->w_scale_;
+  DecodeBoxes(num_boxes, input_boxes, input_anchors, scaler, param->decoded_boxes_);
+  if (param->use_regular_nms_) {
+    *output_num = NmsMultiClassesRegular(num_boxes, num_classes_with_bg, param->decoded_boxes_, input_scores,
+                                         output_boxes, output_classes, output_scores, param);
+  } else {
+    *output_num = NmsMultiClassesFast(num_boxes, num_classes_with_bg, param->decoded_boxes_, input_scores, output_boxes,
+                                      output_classes, output_scores, param);
+  }
+  return NNACL_OK;
+}

mindspore/lite/nnacl/fp32/detection_post_process.h

@@ -0,0 +1,57 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef MINDSPORE_LITE_NNACL_FP32_DETECTION_POST_PROCESS_H_
+#define MINDSPORE_LITE_NNACL_FP32_DETECTION_POST_PROCESS_H_
+
+#include "nnacl/op_base.h"
+#include "nnacl/detection_post_process_parameter.h"
+
+typedef struct {
+  float y;
+  float x;
+  float h;
+  float w;
+} BboxCenter;
+
+typedef struct {
+  float ymin;
+  float xmin;
+  float ymax;
+  float xmax;
+} BboxCorner;
+
+typedef struct {
+  float score;
+  int index;
+} ScoreWithIndex;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void nms_multi_classes_regular();
+
+void nms_multi_classes_fase();
+
+int DetectionPostProcess(const int num_boxes, const int num_classes_with_bg, float *input_boxes, float *input_scores,
+                         float *input_anchors, float *output_boxes, float *output_classes, float *output_scores,
+                         float *output_num, DetectionPostProcessParameter *param);
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // MINDSPORE_LITE_NNACL_FP32_DETECTION_POST_PROCESS_H_

mindspore/lite/schema/ops.fbs

@@ -351,8 +351,8 @@ table DetectionPostProcess {
     NmsIouThreshold: float;
     NmsScoreThreshold: float;
     MaxDetections: long;
-    DetectionsPreClass: long;
-    MaxClassesPreDetection: long;
+    DetectionsPerClass: long;
+    MaxClassesPerDetection: long;
     NumClasses: long;
     UseRegularNms: bool;
     OutQuantized: bool;

mindspore/lite/src/ops/detection_post_process.cc

@@ -34,11 +34,11 @@ float DetectionPostProcess::GetNmsScoreThreshold() const {
 int64_t DetectionPostProcess::GetMaxDetections() const {
   return this->primitive_->value.AsDetectionPostProcess()->MaxDetections;
 }
-int64_t DetectionPostProcess::GetDetectionsPreClass() const {
-  return this->primitive_->value.AsDetectionPostProcess()->DetectionsPreClass;
+int64_t DetectionPostProcess::GetDetectionsPerClass() const {
+  return this->primitive_->value.AsDetectionPostProcess()->DetectionsPerClass;
 }
-int64_t DetectionPostProcess::GetMaxClassesPreDetection() const {
-  return this->primitive_->value.AsDetectionPostProcess()->MaxClassesPreDetection;
+int64_t DetectionPostProcess::GetMaxClassesPerDetection() const {
+  return this->primitive_->value.AsDetectionPostProcess()->MaxClassesPerDetection;
 }
 int64_t DetectionPostProcess::GetNumClasses() const {
   return this->primitive_->value.AsDetectionPostProcess()->NumClasses;
@@ -46,7 +46,6 @@ int64_t DetectionPostProcess::GetNumClasses() const {
 bool DetectionPostProcess::GetUseRegularNms() const {
   return this->primitive_->value.AsDetectionPostProcess()->UseRegularNms;
 }
-
 void DetectionPostProcess::SetFormat(int format) {
   this->primitive_->value.AsDetectionPostProcess()->format = (schema::Format)format;
 }
@@ -72,13 +71,13 @@ void DetectionPostProcess::SetNmsScoreThreshold(float nms_score_threshold) {
   this->primitive_->value.AsDetectionPostProcess()->NmsScoreThreshold = nms_score_threshold;
 }
 void DetectionPostProcess::SetMaxDetections(int64_t max_detections) {
-  this->primitive_->value.AsDetectionPostProcess()->MaxClassesPreDetection = max_detections;
+  this->primitive_->value.AsDetectionPostProcess()->MaxDetections = max_detections;
 }
-void DetectionPostProcess::SetDetectionsPreClass(int64_t detections_pre_class) {
-  this->primitive_->value.AsDetectionPostProcess()->DetectionsPreClass = detections_pre_class;
+void DetectionPostProcess::SetDetectionsPerClass(int64_t detections_per_class) {
+  this->primitive_->value.AsDetectionPostProcess()->DetectionsPerClass = detections_per_class;
 }
-void DetectionPostProcess::SetMaxClassesPreDetection(int64_t max_classes_pre_detection) {
-  this->primitive_->value.AsDetectionPostProcess()->MaxClassesPreDetection = max_classes_pre_detection;
+void DetectionPostProcess::SetMaxClassesPerDetection(int64_t max_classes_per_detection) {
+  this->primitive_->value.AsDetectionPostProcess()->MaxClassesPerDetection = max_classes_per_detection;
 }
 void DetectionPostProcess::SetNumClasses(int64_t num_classes) {
   this->primitive_->value.AsDetectionPostProcess()->NumClasses = num_classes;
@@ -86,6 +85,9 @@ void DetectionPostProcess::SetNumClasses(int64_t num_classes) {
 void DetectionPostProcess::SetUseRegularNms(bool use_regular_nms) {
   this->primitive_->value.AsDetectionPostProcess()->UseRegularNms = use_regular_nms;
 }
+void DetectionPostProcess::SetOutQuantized(bool out_quantized) {
+  this->primitive_->value.AsDetectionPostProcess()->OutQuantized = out_quantized;
+}
 
 #else
 int DetectionPostProcess::UnPackToFlatBuilder(const schema::Primitive *primitive, flatbuffers::FlatBufferBuilder *fbb) {
@@ -98,8 +100,8 @@ int DetectionPostProcess::UnPackToFlatBuilder(const schema::Primitive *primitive
   }
   auto val_offset = schema::CreateDetectionPostProcess(
     *fbb, attr->format(), attr->inputSize(), attr->hScale(), attr->wScale(), attr->xScale(), attr->yScale(),
-    attr->NmsIouThreshold(), attr->NmsScoreThreshold(), attr->MaxDetections(), attr->DetectionsPreClass(),
-    attr->MaxClassesPreDetection(), attr->NumClasses(), attr->UseRegularNms());
+    attr->NmsIouThreshold(), attr->NmsScoreThreshold(), attr->MaxDetections(), attr->DetectionsPerClass(),
+    attr->MaxClassesPerDetection(), attr->NumClasses(), attr->UseRegularNms(), attr->OutQuantized());
   auto prim_offset = schema::CreatePrimitive(*fbb, schema::PrimitiveType_DetectionPostProcess, val_offset.o);
   fbb->Finish(prim_offset);
   return RET_OK;
@@ -121,11 +123,11 @@ float DetectionPostProcess::GetNmsScoreThreshold() const {
 int64_t DetectionPostProcess::GetMaxDetections() const {
   return this->primitive_->value_as_DetectionPostProcess()->MaxDetections();
 }
-int64_t DetectionPostProcess::GetDetectionsPreClass() const {
-  return this->primitive_->value_as_DetectionPostProcess()->DetectionsPreClass();
+int64_t DetectionPostProcess::GetDetectionsPerClass() const {
+  return this->primitive_->value_as_DetectionPostProcess()->DetectionsPerClass();
 }
-int64_t DetectionPostProcess::GetMaxClassesPreDetection() const {
-  return this->primitive_->value_as_DetectionPostProcess()->MaxClassesPreDetection();
+int64_t DetectionPostProcess::GetMaxClassesPerDetection() const {
+  return this->primitive_->value_as_DetectionPostProcess()->MaxClassesPerDetection();
 }
 int64_t DetectionPostProcess::GetNumClasses() const {
   return this->primitive_->value_as_DetectionPostProcess()->NumClasses();
@@ -133,7 +135,67 @@ int64_t DetectionPostProcess::GetNumClasses() const {
 bool DetectionPostProcess::GetUseRegularNms() const {
   return this->primitive_->value_as_DetectionPostProcess()->UseRegularNms();
 }
+bool DetectionPostProcess::GetOutQuantized() const {
+  return this->primitive_->value_as_DetectionPostProcess()->OutQuantized();
+}
 
 #endif
+namespace {
+constexpr int kDetectionPostProcessOutputNum = 4;
+constexpr int kDetectionPostProcessInputNum = 3;
+}  // namespace
+int DetectionPostProcess::InferShape(std::vector<lite::tensor::Tensor *> inputs_,
+                                     std::vector<lite::tensor::Tensor *> outputs_) {
+  if (outputs_.size() != kDetectionPostProcessOutputNum || inputs_.size() != kDetectionPostProcessInputNum) {
+    MS_LOG(ERROR) << "Invalid output/input size! output size: " << outputs_.size() << ",input size: " << inputs_.size();
+    return RET_PARAM_INVALID;
+  }
+  auto boxes = inputs_.at(0);
+  MS_ASSERT(boxes != nullptr);
+  auto scores = inputs_.at(1);
+  MS_ASSERT(scores != nullptr);
+  auto anchors = inputs_.at(2);
+  MS_ASSERT(anchors != nullptr);
+
+  const auto input_box_shape = boxes->shape();
+  const auto input_scores_shape = scores->shape();
+  const auto input_anchors_shape = anchors->shape();
+  MS_ASSERT(input_scores_shape[2] >= GetNumClasses());
+  MS_ASSERT(input_scores_shape[2] - GetNumClasses() <= 1);
+  MS_ASSERT(input_box_shape[1] = input_scores_shape[1]);
+  MS_ASSERT(input_box_shape[1] = input_anchors_shape[0]);
+
+  auto detected_boxes = outputs_.at(0);
+  MS_ASSERT(detected_boxes != nullptr);
+  auto detected_classes = outputs_.at(1);
+  MS_ASSERT(detected_classes != nullptr);
+  auto detected_scores = outputs_.at(2);
+  MS_ASSERT(detected_scores != nullptr);
+  auto num_det = outputs_.at(3);
+  MS_ASSERT(num_det != nullptr);
+
+  detected_boxes->SetFormat(boxes->GetFormat());
+  detected_boxes->set_data_type(boxes->data_type());
+  detected_classes->SetFormat(boxes->GetFormat());
+  detected_classes->set_data_type(boxes->data_type());
+  detected_scores->SetFormat(boxes->GetFormat());
+  detected_scores->set_data_type(boxes->data_type());
+  num_det->SetFormat(boxes->GetFormat());
+  num_det->set_data_type(boxes->data_type());
+  if (!GetInferFlag()) {
+    return RET_OK;
+  }
+  const auto max_detections = GetMaxDetections();
+  const auto max_classes_per_detection = GetMaxClassesPerDetection();
+  const auto num_detected_boxes = static_cast<int>(max_detections * max_classes_per_detection);
+  const std::vector<int> box_shape{1, num_detected_boxes, 4};
+  const std::vector<int> class_shape{1, num_detected_boxes};
+  const std::vector<int> num_shape{1};
+  detected_boxes->set_shape(box_shape);
+  detected_classes->set_shape(class_shape);
+  detected_scores->set_shape(class_shape);
+  num_det->set_shape(num_shape);
+  return RET_OK;
+}
 }  // namespace lite
 }  // namespace mindspore

mindspore/lite/src/ops/detection_post_process.h

@@ -40,15 +40,17 @@ class DetectionPostProcess : public PrimitiveC {
   void SetNmsIouThreshold(float nms_iou_threshold);
   void SetNmsScoreThreshold(float nms_score_threshold);
   void SetMaxDetections(int64_t max_detections);
-  void SetDetectionsPreClass(int64_t detections_pre_class);
-  void SetMaxClassesPreDetection(int64_t max_classes_pre_detection);
+  void SetDetectionsPerClass(int64_t detections_per_class);
+  void SetMaxClassesPerDetection(int64_t max_classes_per_detection);
   void SetNumClasses(int64_t num_classes);
   void SetUseRegularNms(bool use_regular_nms);
+  void SetOutQuantized(bool out_quantized);
 #else
   DetectionPostProcess() = default;
 
   int UnPackToFlatBuilder(const schema::Primitive *primitive, flatbuffers::FlatBufferBuilder *fbb) override;
 #endif
+  int InferShape(std::vector<lite::tensor::Tensor *> inputs_, std::vector<lite::tensor::Tensor *> outputs_) override;
   int GetFormat() const;
   int GetInputSize() const;
   float GetHScale() const;
@@ -58,11 +60,13 @@ class DetectionPostProcess : public PrimitiveC {
   float GetNmsIouThreshold() const;
   float GetNmsScoreThreshold() const;
   int64_t GetMaxDetections() const;
-  int64_t GetDetectionsPreClass() const;
-  int64_t GetMaxClassesPreDetection() const;
+  int64_t GetDetectionsPerClass() const;
+  int64_t GetMaxClassesPerDetection() const;
   int64_t GetNumClasses() const;
   bool GetUseRegularNms() const;
+  bool GetOutQuantized() const;
 };
+
 }  // namespace lite
 }  // namespace mindspore
 

mindspore/lite/src/populate_parameter.cc

@@ -113,6 +113,7 @@
 #include "src/ops/round.h"
 #include "src/ops/sparse_to_dense.h"
 #include "src/ops/l2_norm.h"
+#include "src/ops/detection_post_process.h"
 #include "nnacl/op_base.h"
 #include "nnacl/fp32/arg_min_max.h"
 #include "nnacl/fp32/cast.h"
@@ -171,6 +172,7 @@
 #include "nnacl/leaky_relu_parameter.h"
 #include "nnacl/sparse_to_dense.h"
 #include "nnacl/l2_norm_parameter.h"
+#include "nnacl/detection_post_process_parameter.h"
 
 namespace mindspore::kernel {
 
@@ -1517,20 +1519,17 @@ OpParameter *PopulateEluParameter(const mindspore::lite::PrimitiveC *primitive)
 
 OpParameter *PopulateL2NormParameter(
     const mindspore::lite::PrimitiveC *primitive) {
-  L2NormParameter *l2_norm_parameter =
-      reinterpret_cast<L2NormParameter *>(malloc(sizeof(L2NormParameter)));
+  L2NormParameter *l2_norm_parameter = reinterpret_cast<L2NormParameter *>(malloc(sizeof(L2NormParameter)));
   if (l2_norm_parameter == nullptr) {
     MS_LOG(ERROR) << "malloc L2NormParameter failed.";
     return nullptr;
   }
   memset(l2_norm_parameter, 0, sizeof(L2NormParameter));
   l2_norm_parameter->op_parameter_.type_ = primitive->Type();
-  auto param = reinterpret_cast<mindspore::lite::L2Norm *>(
-      const_cast<mindspore::lite::PrimitiveC *>(primitive));
+  auto param = reinterpret_cast<mindspore::lite::L2Norm *>(const_cast<mindspore::lite::PrimitiveC *>(primitive));
   auto axis_vec = param->GetAxis();
   l2_norm_parameter->axis_num_ = axis_vec.size();
-  l2_norm_parameter->axis_ =
-      reinterpret_cast<int *>(malloc(axis_vec.size() * sizeof(int)));
+  l2_norm_parameter->axis_ = reinterpret_cast<int *>(malloc(axis_vec.size() * sizeof(int)));
   for (size_t i = 0; i < axis_vec.size(); i++) {
     l2_norm_parameter->axis_[i] = axis_vec[i];
   }
@@ -1542,6 +1541,31 @@ OpParameter *PopulateL2NormParameter(
   return reinterpret_cast<OpParameter *>(l2_norm_parameter);
 }
 
+OpParameter *PopulateDetectionPostProcessParameter(const mindspore::lite::PrimitiveC *primitive) {
+  DetectionPostProcessParameter *detection_post_process_parameter =
+    reinterpret_cast<DetectionPostProcessParameter *>(malloc(sizeof(DetectionPostProcessParameter)));
+  if (detection_post_process_parameter == nullptr) {
+    MS_LOG(ERROR) << "malloc EluParameter failed.";
+    return nullptr;
+  }
+  memset(detection_post_process_parameter, 0, sizeof(DetectionPostProcessParameter));
+  detection_post_process_parameter->op_parameter_.type_ = primitive->Type();
+  auto param =
+    reinterpret_cast<mindspore::lite::DetectionPostProcess *>(const_cast<mindspore::lite::PrimitiveC *>(primitive));
+  detection_post_process_parameter->h_scale_ = param->GetHScale();
+  detection_post_process_parameter->w_scale_ = param->GetWScale();
+  detection_post_process_parameter->x_scale_ = param->GetXScale();
+  detection_post_process_parameter->y_scale_ = param->GetYScale();
+  detection_post_process_parameter->nms_iou_threshold_ = param->GetNmsIouThreshold();
+  detection_post_process_parameter->nms_score_threshold_ = param->GetNmsScoreThreshold();
+  detection_post_process_parameter->max_detections_ = param->GetMaxDetections();
+  detection_post_process_parameter->detections_per_class_ = param->GetDetectionsPerClass();
+  detection_post_process_parameter->max_classes_per_detection_ = param->GetMaxClassesPerDetection();
+  detection_post_process_parameter->num_classes_ = param->GetNumClasses();
+  detection_post_process_parameter->use_regular_nms_ = param->GetUseRegularNms();
+  return reinterpret_cast<OpParameter *>(detection_post_process_parameter);
+}
+
 PopulateParameterRegistry::PopulateParameterRegistry() {
   populate_parameter_funcs_[schema::PrimitiveType_SparseToDense] = PopulateSparseToDenseParameter;
   populate_parameter_funcs_[schema::PrimitiveType_SoftMax] = PopulateSoftmaxParameter;
@@ -1640,6 +1664,7 @@ PopulateParameterRegistry::PopulateParameterRegistry() {
   populate_parameter_funcs_[schema::PrimitiveType_EmbeddingLookup] = PopulateEmbeddingLookupParameter;
   populate_parameter_funcs_[schema::PrimitiveType_Elu] = PopulateEluParameter;
   populate_parameter_funcs_[schema::PrimitiveType_L2Norm] = PopulateL2NormParameter;
+  populate_parameter_funcs_[schema::PrimitiveType_DetectionPostProcess] = PopulateDetectionPostProcessParameter;
 }
 
 PopulateParameterRegistry *PopulateParameterRegistry::GetInstance() {

mindspore/lite/src/runtime/kernel/arm/fp32/detection_post_process.cc

@@ -0,0 +1,100 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include "src/runtime/kernel/arm/fp32/detection_post_process.h"
+#include <vector>
+#include "schema/model_generated.h"
+#include "src/kernel_registry.h"
+#include "include/errorcode.h"
+
+using mindspore::kernel::KERNEL_ARCH::kCPU;
+using mindspore::lite::KernelRegistrar;
+using mindspore::lite::RET_ERROR;
+using mindspore::lite::RET_OK;
+using mindspore::schema::PrimitiveType_DetectionPostProcess;
+
+namespace mindspore::kernel {
+int DetectionPostProcessCPUKernel::Init() { return RET_OK; }
+
+int DetectionPostProcessCPUKernel::ReSize() { return RET_OK; }
+
+int DetectionPostProcessCPUKernel::Run() {
+  auto prepare_ret = Prepare();
+  if (prepare_ret != RET_OK) {
+    MS_LOG(ERROR) << "Prepare fail!ret: " << prepare_ret;
+    return prepare_ret;
+  }
+  auto input_boxes = reinterpret_cast<float *>(in_tensors_.at(0)->Data());
+  auto input_scores = reinterpret_cast<float *>(in_tensors_.at(1)->Data());
+  auto input_anchors = reinterpret_cast<float *>(in_tensors_.at(2)->Data());
+
+  // output_classes and output_num use float type now
+  auto output_boxes = reinterpret_cast<float *>(out_tensors_.at(0)->Data());
+  auto output_classes = reinterpret_cast<float *>(out_tensors_.at(1)->Data());
+  auto output_scores = reinterpret_cast<float *>(out_tensors_.at(2)->Data());
+  auto output_num = reinterpret_cast<float *>(out_tensors_.at(3)->Data());
+
+  MS_ASSERT(context_->allocator != nullptr);
+  const int num_boxes = in_tensors_.at(0)->shape()[1];
+  const int num_classes_with_bg = in_tensors_.at(1)->shape()[2];
+  DetectionPostProcessParameter *parameter = reinterpret_cast<DetectionPostProcessParameter *>(op_parameter_);
+  parameter->decoded_boxes_ = context_->allocator->Malloc(num_boxes * 4 * sizeof(float));
+  parameter->nms_candidate_ = context_->allocator->Malloc(num_boxes * sizeof(uint8_t));
+  parameter->selected_ = context_->allocator->Malloc(num_boxes * sizeof(int));
+  parameter->score_with_class_ = context_->allocator->Malloc(num_boxes * sizeof(ScoreWithIndex));
+  if (parameter->use_regular_nms_) {
+    parameter->score_with_class_all_ =
+      context_->allocator->Malloc((num_boxes + parameter->max_detections_) * sizeof(ScoreWithIndex));
+  } else {
+    parameter->score_with_class_all_ =
+      context_->allocator->Malloc((num_boxes * parameter->num_classes_) * sizeof(ScoreWithIndex));
+  }
+  DetectionPostProcess(num_boxes, num_classes_with_bg, input_boxes, input_scores, input_anchors, output_boxes,
+                       output_classes, output_scores, output_num, parameter);
+  context_->allocator->Free(parameter->decoded_boxes_);
+  context_->allocator->Free(parameter->nms_candidate_);
+  context_->allocator->Free(parameter->selected_);
+  context_->allocator->Free(parameter->score_with_class_);
+  context_->allocator->Free(parameter->score_with_class_all_);
+  return RET_OK;
+}
+
+kernel::LiteKernel *CpuDetectionPostProcessFp32KernelCreator(const std::vector<lite::tensor::Tensor *> &inputs,
+                                                             const std::vector<lite::tensor::Tensor *> &outputs,
+                                                             OpParameter *opParameter, const lite::Context *ctx,
+                                                             const kernel::KernelKey &desc,
+                                                             const mindspore::lite::PrimitiveC *primitive) {
+  if (opParameter == nullptr) {
+    MS_LOG(ERROR) << "Create kernel failed, opParameter is nullptr, type: PrimitiveType_DetectionPostProcess. ";
+    return nullptr;
+  }
+  MS_ASSERT(desc.type == schema::PrimitiveType_DetectionPostProcess);
+  auto *kernel = new (std::nothrow) DetectionPostProcessCPUKernel(opParameter, inputs, outputs, ctx, primitive);
+  if (kernel == nullptr) {
+    MS_LOG(ERROR) << "new DetectionPostProcessCPUKernel fail!";
+    return nullptr;
+  }
+  auto ret = kernel->Init();
+  if (ret != RET_OK) {
+    MS_LOG(ERROR) << "Init kernel failed, name: " << opParameter->name_ << ", type: "
+                  << schema::EnumNamePrimitiveType(static_cast<schema::PrimitiveType>(opParameter->type_));
+    delete kernel;
+    return nullptr;
+  }
+  return kernel;
+}
+
+REG_KERNEL(kCPU, kNumberTypeFloat32, PrimitiveType_DetectionPostProcess, CpuDetectionPostProcessFp32KernelCreator)
+}  // namespace mindspore::kernel

mindspore/lite/src/runtime/kernel/arm/fp32/detection_post_process.h

@@ -0,0 +1,46 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP32_DETECTION_POST_PROCESS_H_
+#define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP32_DETECTION_POST_PROCESS_H_
+
+#include <vector>
+#include "src/lite_kernel.h"
+#include "include/context.h"
+#include "nnacl/fp32/detection_post_process.h"
+
+using mindspore::lite::Context;
+
+namespace mindspore::kernel {
+class DetectionPostProcessCPUKernel : public LiteKernel {
+ public:
+  DetectionPostProcessCPUKernel(OpParameter *parameter, const std::vector<lite::tensor::Tensor *> &inputs,
+                                const std::vector<lite::tensor::Tensor *> &outputs, const lite::Context *ctx,
+                                const mindspore::lite::PrimitiveC *primitive)
+      : LiteKernel(parameter, inputs, outputs, ctx, primitive) {
+    param_ = reinterpret_cast<DetectionPostProcessCPUKernel *>(parameter);
+  }
+  ~DetectionPostProcessCPUKernel() override = default;
+
+  int Init() override;
+  int ReSize() override;
+  int Run() override;
+
+ private:
+  DetectionPostProcessCPUKernel *param_;
+};
+}  // namespace mindspore::kernel
+#endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP32_DETECTION_POST_PROCESS_H_

mindspore/lite/test/ut/src/runtime/kernel/arm/fp32/detection_post_process_test.cc

@@ -0,0 +1,159 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "mindspore/core/utils/log_adapter.h"
+#include "common/common_test.h"
+#include "mindspore/lite/src/runtime/kernel/arm/fp32/detection_post_process.h"
+#include "src/kernel_registry.h"
+#include "src/lite_kernel.h"
+#include "src/common/file_utils.h"
+
+namespace mindspore {
+class TestDetectionPostProcessFp32 : public mindspore::CommonTest {
+ public:
+  TestDetectionPostProcessFp32() {}
+};
+
+void DetectionPostProcessTestInit(std::vector<lite::tensor::Tensor *> *inputs_,
+                                  std::vector<lite::tensor::Tensor *> *outputs_, DetectionPostProcessParameter *param) {
+  std::string input_boxes_path = "./test_data/detectionPostProcess/input_boxes.bin";
+  size_t input_boxes_size;
+  auto input_boxes_data =
+    reinterpret_cast<float *>(mindspore::lite::ReadFile(input_boxes_path.c_str(), &input_boxes_size));
+  auto *input_boxes = new lite::tensor::Tensor;
+  input_boxes->set_data_type(kNumberTypeFloat32);
+  input_boxes->SetFormat(schema::Format_NHWC);
+  input_boxes->set_shape({1, 1917, 4});
+  input_boxes->MallocData();
+  memcpy(input_boxes->Data(), input_boxes_data, input_boxes_size);
+  inputs_->push_back(input_boxes);
+
+  std::string input_scores_path = "./test_data/detectionPostProcess/input_scores.bin";
+  size_t input_scores_size;
+  auto input_scores_data =
+    reinterpret_cast<float *>(mindspore::lite::ReadFile(input_scores_path.c_str(), &input_scores_size));
+  auto *input_scores = new lite::tensor::Tensor;
+  input_scores->set_data_type(kNumberTypeFloat32);
+  input_scores->SetFormat(schema::Format_NHWC);
+  input_scores->set_shape({1, 1917, 91});
+  input_scores->MallocData();
+  memcpy(input_scores->Data(), input_scores_data, input_scores_size);
+  inputs_->push_back(input_scores);
+
+  std::string input_anchors_path = "./test_data/detectionPostProcess/input_anchors.bin";
+  size_t input_anchors_size;
+  auto input_anchors_data =
+    reinterpret_cast<float *>(mindspore::lite::ReadFile(input_anchors_path.c_str(), &input_anchors_size));
+  auto *input_anchors = new lite::tensor::Tensor;
+  input_anchors->set_data_type(kNumberTypeFloat32);
+  input_anchors->SetFormat(schema::Format_NHWC);
+  input_anchors->set_shape({1917, 4});
+  input_anchors->MallocData();
+  memcpy(input_anchors->Data(), input_anchors_data, input_anchors_size);
+  inputs_->push_back(input_anchors);
+
+  auto *output_boxes = new lite::tensor::Tensor;
+  output_boxes->set_data_type(kNumberTypeFloat32);
+  output_boxes->set_shape({1, 10, 4});
+  output_boxes->SetFormat(schema::Format_NHWC);
+  output_boxes->MallocData();
+  memset(output_boxes->Data(), 0, output_boxes->ElementsNum() * sizeof(float));
+
+  auto *output_classes = new lite::tensor::Tensor;
+  output_classes->set_data_type(kNumberTypeFloat32);
+  output_classes->set_shape({1, 10});
+  output_classes->SetFormat(schema::Format_NHWC);
+  output_classes->MallocData();
+  memset(output_classes->Data(), 0, output_classes->ElementsNum() * sizeof(float));
+
+  auto *output_scores = new lite::tensor::Tensor;
+  output_scores->set_data_type(kNumberTypeFloat32);
+  output_scores->set_shape({1, 10});
+  output_scores->SetFormat(schema::Format_NHWC);
+  output_scores->MallocData();
+  memset(output_scores->Data(), 0, output_scores->ElementsNum() * sizeof(float));
+
+  auto *output_num_det = new lite::tensor::Tensor;
+  output_num_det->set_data_type(kNumberTypeFloat32);
+  output_num_det->set_shape({1});
+  output_num_det->SetFormat(schema::Format_NHWC);
+  output_num_det->MallocData();
+  memset(output_num_det->Data(), 0, output_num_det->ElementsNum() * sizeof(float));
+
+  outputs_->push_back(output_boxes);
+  outputs_->push_back(output_classes);
+  outputs_->push_back(output_scores);
+  outputs_->push_back(output_num_det);
+
+  param->h_scale_ = 5;
+  param->w_scale_ = 5;
+  param->x_scale_ = 10;
+  param->y_scale_ = 10;
+  param->nms_iou_threshold_ = 0.6;
+  param->nms_score_threshold_ = 1e-8;
+  param->max_detections_ = 10;
+  param->detections_per_class_ = 100;
+  param->max_classes_per_detection_ = 1;
+  param->num_classes_ = 90;
+  param->use_regular_nms_ = false;
+  param->out_quantized_ = true;
+}
+
+TEST_F(TestDetectionPostProcessFp32, Fast) {
+  std::vector<lite::tensor::Tensor *> inputs_;
+  std::vector<lite::tensor::Tensor *> outputs_;
+  auto param = new DetectionPostProcessParameter();
+  DetectionPostProcessTestInit(&inputs_, &outputs_, param);
+  auto ctx = new lite::Context;
+  ctx->thread_num_ = 1;
+  kernel::DetectionPostProcessCPUKernel *op =
+    new kernel::DetectionPostProcessCPUKernel(reinterpret_cast<OpParameter *>(param), inputs_, outputs_, ctx, nullptr);
+  op->Init();
+  op->Run();
+
+  float *output_boxes = reinterpret_cast<float *>(outputs_[0]->Data());
+  size_t output_boxes_size;
+  std::string output_boxes_path = "./test_data/detectionPostProcess/output_0.bin";
+  auto correct_boxes =
+    reinterpret_cast<float *>(mindspore::lite::ReadFile(output_boxes_path.c_str(), &output_boxes_size));
+  CompareOutputData(output_boxes, correct_boxes, outputs_[0]->ElementsNum(), 0.0001);
+
+  float *output_classes = reinterpret_cast<float *>(outputs_[1]->Data());
+  size_t output_classes_size;
+  std::string output_classes_path = "./test_data/detectionPostProcess/output_1.bin";
+  auto correct_classes =
+    reinterpret_cast<float *>(mindspore::lite::ReadFile(output_classes_path.c_str(), &output_classes_size));
+  CompareOutputData(output_classes, correct_classes, outputs_[1]->ElementsNum(), 0.0001);
+
+  float *output_scores = reinterpret_cast<float *>(outputs_[2]->Data());
+  size_t output_scores_size;
+  std::string output_scores_path = "./test_data/detectionPostProcess/output_2.bin";
+  auto correct_scores =
+    reinterpret_cast<float *>(mindspore::lite::ReadFile(output_scores_path.c_str(), &output_scores_size));
+  CompareOutputData(output_scores, correct_scores, outputs_[2]->ElementsNum(), 0.0001);
+
+  float *output_num_det = reinterpret_cast<float *>(outputs_[3]->Data());
+  size_t output_num_det_size;
+  std::string output_num_det_path = "./test_data/detectionPostProcess/output_3.bin";
+  auto correct_num_det =
+    reinterpret_cast<float *>(mindspore::lite::ReadFile(output_num_det_path.c_str(), &output_num_det_size));
+  CompareOutputData(output_num_det, correct_num_det, outputs_[3]->ElementsNum(), 0.0001);
+
+  delete op;
+  for (auto t : inputs_) delete t;
+  for (auto t : outputs_) delete t;
+}
+}  // namespace mindspore

mindspore/lite/tools/converter/parser/tflite/tflite_custom_parser.cc

@@ -57,11 +57,11 @@ STATUS TfliteCustomParser::Parse(const std::unique_ptr<tflite::OperatorT> &tflit
   attr->NmsScoreThreshold = attr_map["nms_score_threshold"].AsFloat();
   attr->MaxDetections = attr_map["max_detections"].AsInt32();
   if (attr_map["detections_per_class"].IsNull()) {
-    attr->DetectionsPreClass = 100;
+    attr->DetectionsPerClass = 100;
   } else {
-    attr->DetectionsPreClass = attr_map["detections_per_class"].AsInt32();
+    attr->DetectionsPerClass = attr_map["detections_per_class"].AsInt32();
   }
-  attr->MaxClassesPreDetection = attr_map["max_classes_per_detection"].AsInt32();
+  attr->MaxClassesPerDetection = attr_map["max_classes_per_detection"].AsInt32();
   attr->NumClasses = attr_map["num_classes"].AsInt32();
   if (attr_map["use_regular_nms"].IsNull()) {
     attr->UseRegularNms = false;