!5807 detection_post_process op support quantized anchors tensor

Merge pull request !5807 from wangzhe/master

mindspore/lite/nnacl/detection_post_process_parameter.h

@@ -33,6 +33,8 @@ typedef struct DetectionPostProcessParameter {
   bool use_regular_nms_;
   bool out_quantized_;
 
+  float *anchors_;
+
   void *decoded_boxes_;
   void *nms_candidate_;
   void *selected_;

mindspore/lite/nnacl/fp32/detection_post_process.c

@@ -44,15 +44,15 @@ float IntersectionOverUnion(const BboxCorner *a, const BboxCorner *b) {
   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);
+  return inter / (area_a + area_b - inter);
 }
 
 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);
+    BboxCenter *box = (BboxCenter *)(input_boxes) + i;
+    BboxCenter *anchor = (BboxCenter *)(anchors) + i;
+    BboxCorner *decoded_box = (BboxCorner *)(decoded_boxes) + i;
     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;
@@ -137,7 +137,7 @@ int NmsMultiClassesRegular(const int num_boxes, const int num_classes_with_bg, c
       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;;
+      output_scores[i] = score_with_index_all[i].score;
     } else {
       ((BboxCorner *)(output_boxes) + i)->ymin = 0;
       ((BboxCorner *)(output_boxes) + i)->xmin = 0;

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

@@ -26,7 +26,32 @@ using mindspore::lite::RET_OK;
 using mindspore::schema::PrimitiveType_DetectionPostProcess;
 
 namespace mindspore::kernel {
-int DetectionPostProcessCPUKernel::Init() { return RET_OK; }
+int DetectionPostProcessCPUKernel::Init() {
+  MS_ASSERT(context_->allocator != nullptr);
+  auto anchor_tensor = in_tensors_.at(2);
+  DetectionPostProcessParameter *parameter = reinterpret_cast<DetectionPostProcessParameter *>(op_parameter_);
+  if (anchor_tensor->data_type() == kNumberTypeUInt8) {
+    const auto quant_params = anchor_tensor->GetQuantParams();
+    const double scale = quant_params.at(0).scale;
+    const int32_t zp = quant_params.at(0).zeroPoint;
+    auto anchor_uint8 = reinterpret_cast<uint8_t *>(anchor_tensor->Data());
+    auto anchor_fp32 =
+      reinterpret_cast<float *>(context_->allocator->Malloc(anchor_tensor->ElementsNum() * sizeof(float)));
+    for (int i = 0; i < anchor_tensor->ElementsNum(); ++i) {
+      *(anchor_fp32 + i) = static_cast<float>((static_cast<int>(anchor_uint8[i]) - zp) * scale);
+    }
+    parameter->anchors_ = anchor_fp32;
+  } else if (anchor_tensor->data_type() == kNumberTypeFloat32) {
+    auto anchor_fp32 = reinterpret_cast<float *>(anchor_tensor->Data());
+    for (int i = 0; i < anchor_tensor->ElementsNum(); ++i) {
+      parameter->anchors_[i] = anchor_fp32[i];
+    }
+  } else {
+    MS_LOG(ERROR) << "unsupported anchor data type " << anchor_tensor->data_type();
+    return RET_ERROR;
+  }
+  return RET_OK;
+}
 
 int DetectionPostProcessCPUKernel::ReSize() { return RET_OK; }
 
@@ -38,7 +63,6 @@ int DetectionPostProcessCPUKernel::Run() {
   }
   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());
@@ -61,7 +85,7 @@ int DetectionPostProcessCPUKernel::Run() {
     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,
+  DetectionPostProcess(num_boxes, num_classes_with_bg, input_boxes, input_scores, parameter->anchors_, output_boxes,
                        output_classes, output_scores, output_num, parameter);
   context_->allocator->Free(parameter->decoded_boxes_);
   context_->allocator->Free(parameter->nms_candidate_);

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

@@ -56,9 +56,13 @@ void DetectionPostProcessTestInit(std::vector<lite::tensor::Tensor *> *inputs_,
   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));
+    reinterpret_cast<uint8_t *>(mindspore::lite::ReadFile(input_anchors_path.c_str(), &input_anchors_size));
   auto *input_anchors = new lite::tensor::Tensor;
-  input_anchors->set_data_type(kNumberTypeFloat32);
+  lite::tensor::QuantArg quant_arg;
+  quant_arg.zeroPoint = 0;
+  quant_arg.scale = 0.00645306;
+  input_anchors->AddQuantParam(quant_arg);
+  input_anchors->set_data_type(kNumberTypeUInt8);
   input_anchors->SetFormat(schema::Format_NHWC);
   input_anchors->set_shape({1917, 4});
   input_anchors->MallocData();