strided_slice_with_axes

mindspore/lite/nnacl/strided_slice.c

@@ -104,8 +104,12 @@ int DoStridedSlice(const void *in_data, void *out_data, StridedSliceParameter *p
                                   dim3 * dim_offset[3] + dim4 * dim_offset[4] + dim5;
               if (param->data_type == kDataTypeFloat) {
                 *((float *)out_data + out_offset) = *((float *)in_data + in_offset);
-              } else {
+              } else if (param->data_type == kDataTypeInt8) {
                 *((int8_t *)out_data + out_offset) = *((int8_t *)in_data + in_offset);
+              } else if (param->data_type == kDataTypeInt) {
+                *((int32_t *)out_data + out_offset) = *((int32_t *)in_data + in_offset);
+              } else {
+                return NNACL_ERR;
               }
               out_offset++;
             }

mindspore/lite/src/ops/constant_of_shape.cc

@@ -87,7 +87,7 @@ int ConstantOfShape::InferShape(std::vector<Tensor *> inputs_, std::vector<Tenso
   }
   auto in_data = reinterpret_cast<int *>(in_tensor->data_c());
   if (in_data == nullptr) {
-    MS_LOG(ERROR) << "Input data is nullptr";
+    MS_LOG(INFO) << "Input data is nullptr. Input tensor has not been calculated out yet.";
     return RET_INFER_INVALID;
   }
   int size = in_tensor->ElementsNum();

mindspore/lite/src/ops/populate/populate_register.h

@@ -50,6 +50,7 @@ class Registry {
   }
 };
 OpParameter *PopulateArithmetic(const mindspore::lite::PrimitiveC *primitive);
+OpParameter *PopulateStridedSliceParameter(const mindspore::lite::PrimitiveC *primitive);
 }  // namespace lite
 }  // namespace mindspore
 #endif

mindspore/lite/src/ops/populate/strided_slice_populate.cc

@@ -41,6 +41,7 @@ OpParameter *PopulateStridedSliceParameter(const mindspore::lite::PrimitiveC *pr
   memcpy(strided_slice_param->strides_, (stride.data()), stride.size() * sizeof(int));
   auto in_shape = ((lite::StridedSlice *)primitive)->GetInShape();
   memcpy(strided_slice_param->in_shape_, (in_shape.data()), in_shape.size() * sizeof(int));
+  strided_slice_param->in_shape_length_ = static_cast<int>(in_shape.size());
   return reinterpret_cast<OpParameter *>(strided_slice_param);
 }
 

mindspore/lite/src/ops/strided_slice.cc

@@ -15,6 +15,7 @@
  */
 
 #include "src/ops/strided_slice.h"
+#include <algorithm>
 
 #ifndef PRIMITIVE_WRITEABLE
 #include "src/ops/ops_register.h"
@@ -172,7 +173,8 @@ Registry StridedSliceRegistry(schema::PrimitiveType_StridedSlice, StridedSliceCr
 namespace {
 constexpr size_t kStridedSliceOutputNum = 1;
 constexpr size_t kStridedSliceInputNum = 1;
-constexpr size_t kStridedSliceMultiInputNum = 4;
+constexpr size_t kStridedSliceMultiInputNumMin = 3;
+constexpr size_t kStridedSliceMultiInputNumMax = 5;
 }  // namespace
 
 void StridedSlice::ApplyNewAxisMask() {
@@ -251,13 +253,91 @@ void StridedSlice::TransIndexToPositive() {
   }
 }
 
+int StridedSlice::HandleAxesInputExist(const std::vector<lite::Tensor *> &inputs) {
+  // when axes input exist:
+  // input order: data, begin, end, axes(opt), stride(opt)
+  auto input_tensor = inputs.at(0);
+  MS_ASSERT(input_tensor != nullptr);
+  auto begin_tensor = inputs.at(1);
+  MS_ASSERT(begin_tensor != nullptr);
+  int *begin_data = reinterpret_cast<int *>(begin_tensor->MutableData());
+  auto end_tensor = inputs.at(2);
+  MS_ASSERT(end_tensor != nullptr);
+  int *end_data = reinterpret_cast<int *>(end_tensor->MutableData());
+  if (begin_data == nullptr || end_data == nullptr) {
+    return RET_INFER_ERR;
+  }
+  // when input contains axes, begins, ends, strides will be expand to the same length as input rank
+  ndim_ = static_cast<int>(input_tensor->shape().size());
+  int begin_ndim = begin_tensor->ElementsNum();
+
+  int *axes_data = nullptr;
+  auto axes_tensor = inputs.at(3);
+  if (axes_tensor->ElementsNum() != 0) {
+    MS_ASSERT(axes_tensor->ElementsNum() == begin_ndim);
+    axes_data = reinterpret_cast<int *>(axes_tensor->MutableData());
+    if (axes_data == nullptr) {
+      return RET_INFER_ERR;
+    }
+  }
+
+  int *stride_data = nullptr;
+  auto stride_tensor = inputs.at(4);
+  if (stride_tensor->ElementsNum() != 0) {
+    MS_ASSERT(stride_tensor->ElementsNum() == begin_ndim);
+    stride_data = reinterpret_cast<int *>(stride_tensor->MutableData());
+    if (stride_data == nullptr) {
+      return RET_INFER_ERR;
+    }
+  }
+
+  std::vector<int> axes;
+  if (axes_data == nullptr) {
+    for (int i = 0; i < begin_ndim; ++i) {
+      axes[i] = i;
+    }
+  } else {
+    axes.assign(axes_data, axes_data + begin_ndim);
+    for (int i = 0; i < begin_ndim; ++i) {
+      if (axes[i] < 0) {
+        axes[i] += ndim_;
+      }
+    }
+  }
+
+  in_shape_.assign(ndim_, 0);
+  begins_.assign(ndim_, 0);
+  ends_.assign(ndim_, 0);
+  strides_.assign(ndim_, 0);
+  auto input_shape = input_tensor->shape();
+  for (int i = 0; i < ndim_; ++i) {
+    in_shape_[i] = input_shape.at(i);
+  }
+  for (int i = 0; i < ndim_; ++i) {
+    auto axes_it = std::find(axes.begin(), axes.end(), i);
+    if (axes_it != axes.end()) {
+      auto axis = axes_it - axes.begin();
+      // begins or ends exceed limit will be set to limit
+      begins_[i] = std::max(std::min(begin_data[axis], input_shape[i] - 1), -input_shape[i]);
+      ends_[i] = std::max(std::min(end_data[axis], input_shape[i]), -input_shape[i] - 1);
+      strides_[i] = stride_data[axis];
+    } else {
+      begins_[i] = 0;
+      ends_[i] = input_shape[i];
+      strides_[i] = 1;
+    }
+  }
+  return RET_OK;
+}
+
 int StridedSlice::InferShape(std::vector<lite::Tensor *> inputs, std::vector<lite::Tensor *> outputs) {
   MS_ASSERT(this->primitive_ != nullptr);
   if (outputs.size() != kStridedSliceOutputNum) {
     MS_LOG(ERROR) << "Invalid output size:" << outputs.size();
     return RET_PARAM_INVALID;
   }
-  if (inputs.size() != kStridedSliceInputNum && inputs.size() != kStridedSliceMultiInputNum) {
+  if (inputs.size() != kStridedSliceInputNum &&
+      !(inputs.size() <= kStridedSliceMultiInputNumMax && inputs.size() >= kStridedSliceMultiInputNumMin)) {
     MS_LOG(ERROR) << "Invalid input size " << inputs.size();
     return RET_PARAM_INVALID;
   }
@@ -268,6 +348,10 @@ int StridedSlice::InferShape(std::vector<lite::Tensor *> inputs, std::vector<lit
   auto input_shape = input->shape();
   auto inferflag = GetInferFlag();
 
+  in_shape_.clear();
+  begins_.clear();
+  ends_.clear();
+  strides_.clear();
   if (inputs.size() == kStridedSliceInputNum) {
     ndim_ = static_cast<int>(GetBegin().size());
 
@@ -279,7 +363,9 @@ int StridedSlice::InferShape(std::vector<lite::Tensor *> inputs, std::vector<lit
       ends_.emplace_back((GetEnd())[i]);
       strides_.emplace_back((GetStride())[i]);
     }
-  } else {
+  }
+  if (inputs.size() == 4) {
+    // input order: input, begins, ends, strides.
     auto begin_tensor = inputs.at(1);
     int *begin_data = reinterpret_cast<int *>(begin_tensor->MutableData());
     auto end_tensor = inputs.at(2);
@@ -299,6 +385,13 @@ int StridedSlice::InferShape(std::vector<lite::Tensor *> inputs, std::vector<lit
       strides_.emplace_back(stride_data[i]);
     }
   }
+  if (inputs.size() == 5) {
+    // input order: input, begins, end, axes, strides
+    auto ret = HandleAxesInputExist(inputs);
+    if (ret != RET_OK) {
+      return ret;
+    }
+  }
 
   // set all mask to original input shape
   begins_mask_.resize(ndim_);
@@ -333,7 +426,12 @@ int StridedSlice::InferShape(std::vector<lite::Tensor *> inputs, std::vector<lit
     if (i < ndim_ && new_axis_mask_.at(i)) {
       output_shape.at(i) = 1;
     } else {
-      output_shape.at(i) = (ends_.at(i) - begins_.at(i)) / strides_.at(i);
+      if (strides_.at(i) == 0) {
+        MS_LOG(ERROR) << "strides should not be 0.";
+        return RET_INFER_ERR;
+      }
+      output_shape.at(i) =
+        (ends_.at(i) - begins_.at(i) + strides_.at(i) + (strides_.at(i) < 0 ? 1 : -1)) / strides_.at(i);
     }
   }
 

mindspore/lite/src/ops/strided_slice.h

@@ -81,6 +81,7 @@ class StridedSlice : public PrimitiveC {
   std::vector<bool> new_axis_mask_;
   std::vector<bool> shrink_axis_mask_;
   void TransIndexToPositive();
+  int HandleAxesInputExist(const std::vector<lite::Tensor *> &inputs);
 };
 }  // namespace lite
 }  // namespace mindspore

mindspore/lite/src/runtime/kernel/arm/base/strided_slice.cc

@@ -16,11 +16,11 @@
 
 #include "src/runtime/kernel/arm/base/strided_slice.h"
 #include <vector>
-#include "nnacl/strided_slice.h"
 #include "schema/model_generated.h"
 #include "src/kernel_registry.h"
 #include "include/errorcode.h"
 #include "src/runtime/runtime_api.h"
+#include "src/ops/populate/populate_register.h"
 
 using mindspore::kernel::KERNEL_ARCH::kCPU;
 using mindspore::lite::KernelRegistrar;
@@ -44,16 +44,16 @@ int StridedSliceCPUKernel::Init() {
 }
 
 int StridedSliceCPUKernel::ReSize() {
-  auto input = in_tensors_.at(0);
-  auto parameter = reinterpret_cast<StridedSliceParameter *>(op_parameter_);
-  MS_ASSERT(input);
-  MS_ASSERT(parameter);
-  parameter->data_type = input->data_type() == kNumberTypeInt8 ? kDataTypeInt8 : kDataTypeFloat;
-  auto input_shape = input->shape();
-  for (size_t i = 0; i < input_shape.size(); ++i) {
-    parameter->in_shape_[i] = input_shape[i];
+  if (op_parameter_ != nullptr) {
+    free(op_parameter_);
+    op_parameter_ = nullptr;
+  }
+  op_parameter_ = PopulateStridedSliceParameter(primitive_);
+  if (op_parameter_ == nullptr) {
+    MS_LOG(ERROR) << "Malloc parameter failed";
+    return RET_ERROR;
   }
-  parameter->in_shape_length_ = static_cast<int>(input_shape.size());
+  param_ = reinterpret_cast<StridedSliceParameter *>(op_parameter_);
   return RET_OK;
 }
 
@@ -62,8 +62,7 @@ int StridedSliceCPUKernel::HandleMultiInputs() {
     MS_LOG(ERROR) << "Inputs size should be " << kMultiInputsSize << ", got " << in_tensors_.size();
     return RET_ERROR;
   }
-  auto param = reinterpret_cast<StridedSliceParameter *>(op_parameter_);
-  if (param == nullptr) {
+  if (param_ == nullptr) {
     MS_LOG(ERROR) << "StridedSliceParamater cast nullptr";
     return RET_ERROR;
   }
@@ -74,35 +73,49 @@ int StridedSliceCPUKernel::HandleMultiInputs() {
     MS_LOG(ERROR) << "StridedSlice supports max dimension " << DIMENSION_6D << ", input begins dim is " << axis_num;
     return RET_ERROR;
   }
-  memcpy(param->begins_, begins->MutableData(), axis_num * sizeof(int));
+  memcpy(param_->begins_, begins->MutableData(), axis_num * sizeof(int));
 
   auto ends = in_tensors_.at(kEndsIndex);
   MS_ASSERT(ends != nullptr);
   MS_ASSERT(axis_num == ends->ElementsNum());
-  memcpy(param->ends_, ends->MutableData(), axis_num * sizeof(int));
+  memcpy(param_->ends_, ends->MutableData(), axis_num * sizeof(int));
 
   auto strides = in_tensors_.at(kStridesInex);
   MS_ASSERT(strides != nullptr);
   MS_ASSERT(axis_num == strides->ElementsNum());
-  memcpy(param->strides_, strides->MutableData(), axis_num * sizeof(int));
+  memcpy(param_->strides_, strides->MutableData(), axis_num * sizeof(int));
 
-  param->num_axes_ = axis_num;
+  param_->num_axes_ = axis_num;
   return RET_OK;
 }
 
 int StridedSliceCPUKernel::Run() {
   auto input = in_tensors_.at(0);
-  auto output = out_tensors_.at(0);
   MS_ASSERT(input);
+  switch (input->data_type()) {
+    case kNumberTypeInt8:
+      param_->data_type = kDataTypeInt8;
+      break;
+    case kNumberTypeFloat32:
+      param_->data_type = kDataTypeFloat;
+      break;
+    case kNumberTypeInt32:
+      param_->data_type = kDataTypeInt;
+      break;
+    default:
+      MS_LOG(ERROR) << "Not supported data type: " << input->data_type();
+      return RET_ERROR;
+  }
+  auto output = out_tensors_.at(0);
   MS_ASSERT(output);
+  // inputs order: input, begin, end, stride
   if (in_tensors().size() == kMultiInputsSize) {
     auto ret = HandleMultiInputs();
     if (ret != RET_OK) {
       return ret;
     }
   }
-  auto ret = DoStridedSlice(input->MutableData(), output->MutableData(),
-                            reinterpret_cast<StridedSliceParameter *>(op_parameter_));
+  auto ret = DoStridedSlice(input->MutableData(), output->MutableData(), param_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "StridedSlice error error_code[" << ret << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/base/strided_slice.h

@@ -18,7 +18,7 @@
 #define MINDSPORE_LITE_SRC_BACKEND_ARM_BASE_STRIDED_SLICE_H_
 
 #include <vector>
-
+#include "nnacl/strided_slice.h"
 #include "src/lite_kernel.h"
 
 namespace mindspore::kernel {
@@ -27,7 +27,9 @@ class StridedSliceCPUKernel : public LiteKernel {
   StridedSliceCPUKernel(OpParameter *parameter, const std::vector<lite::Tensor *> &inputs,
                         const std::vector<lite::Tensor *> &outputs, const lite::InnerContext *ctx,
                         const mindspore::lite::PrimitiveC *primitive)
-      : LiteKernel(parameter, inputs, outputs, ctx, primitive) {}
+      : LiteKernel(parameter, inputs, outputs, ctx, primitive) {
+    param_ = reinterpret_cast<StridedSliceParameter *>(parameter);
+  }
   ~StridedSliceCPUKernel() override = default;
 
   int Init() override;
@@ -36,6 +38,9 @@ class StridedSliceCPUKernel : public LiteKernel {
 
  private:
   int HandleMultiInputs();
+
+ private:
+  StridedSliceParameter *param_;
 };
 }  // namespace mindspore::kernel
 

mindspore/lite/test/models_onnx.cfg

@@ -25,9 +25,3 @@ psenet_lite_mbv2.onnx;1,32,32,3
 super-resolution-10.onnx;1,224,224,1
 tinyyolov2-8.onnx;1,416,416,3
 ml_2012_ocr_cn.onnx
-ml_2012_ocr_cn_noLSTM.onnx
-candy-9.onnx
-mosaic-9.onnx
-pointilism-9.onnx
-rain-princess-9.onnx
-udnie-9.onnx

mindspore/lite/test/run_benchmark_nets.sh

@@ -62,7 +62,7 @@ function Run_Converter() {
         if [ $? = 0 ]; then
             converter_result='converter onnx '${model_name}' pass';echo ${converter_result} >> ${run_converter_result_file}
         else
-            converter_result='converter onnx '${model_name}' failed';echo ${converter_result} >> ${run_converter_result_file}
+            converter_result='converter onnx '${model_name}' failed';echo ${converter_result} >> ${run_converter_result_file};return 1
         fi
     done < ${models_onnx_config}
 

mindspore/lite/tools/anf_importer/import_from_meta_graphT.cc

@@ -64,6 +64,9 @@ int AnfImporterFromMetaGraphT::ConverterConstTensor() {
       param_value->set_tensor_addr(tensor_data);
       param_value->set_tensor_size(size);
       parameter->set_default_param(param_value);
+    } else if (std::find(meta_graph_->inputIndex.begin(), meta_graph_->inputIndex.end(), i) ==
+               meta_graph_->inputIndex.end()) {
+      parameter->set_default_param(param_value);
     }
     AddNode(i, parameter);
   }

mindspore/lite/tools/converter/parser/onnx/onnx_model_parser.h

@@ -30,7 +30,7 @@
 #include "securec/include/securec.h"
 #include "tools/converter/model_parser.h"
 #include "tools/converter/parser/onnx/onnx_node_parser_registry.h"
-#include "tools/common/tensor_util.h"
+#include "tools/converter/parser/onnx/onnx_tensor_parser.h"
 #include "proto/onnx.pb.h"
 
 namespace mindspore {
@@ -53,42 +53,38 @@ class OnnxModelParser : public ModelParser {
  private:
   std::vector<int32_t> GetDimsFromOnnxValue(const onnx::ValueInfoProto &onnx_value);
 
-  STATUS SetGraphConstTensor(const onnx::GraphProto &onnx_graph, TensorCache *tensor_cache);
+  STATUS SetGraphConstTensor(const onnx::GraphProto &onnx_graph);
 
-  STATUS SetGraphInputTensor(const onnx::GraphProto &onnx_graph, schema::SubGraphT *graph, TensorCache *tensor_cache);
+  STATUS SetGraphInputTensor(const onnx::GraphProto &onnx_graph, schema::SubGraphT *graph);
 
-  STATUS SetGraphOutputTensor(const onnx::GraphProto &onnx_graph, schema::SubGraphT *graph, TensorCache *tensor_cache);
+  STATUS SetGraphOutputTensor(const onnx::GraphProto &onnx_graph, schema::SubGraphT *graph);
 
-  STATUS AddValueInfo(const onnx::ValueInfoProto &proto, const std::string &name, const Category &type,
-                      TensorCache *tensor_cache, int *index);
+  STATUS AddValueInfo(const onnx::ValueInfoProto &proto, const std::string &name, const Category &type, int *index);
 
-  STATUS AddTensorProto(const onnx::TensorProto &proto, const std::string &name, const Category &type,
-                        TensorCache *tensor_cache, int *index);
+  STATUS AddTensorProto(const onnx::TensorProto &proto, const std::string &name, const Category &type, int *index);
 
   STATUS ParseOnnxNodeToDstOp(const onnx::GraphProto &onnx_graph, const onnx::NodeProto &onnx_node,
-                              schema::CNodeT *dst_op, TensorCache *tensor_cache, const QuantType &quantType,
-                              schema::MetaGraphT *dst_graph);
+                              schema::CNodeT *dst_op, const QuantType &quantType, schema::MetaGraphT *dst_graph);
 
   void ParseOnnxGemmNode(const onnx::GraphProto &onnx_graph, const onnx::NodeProto &onnx_node,
-                         schema::SubGraphT *sub_graph, schema::MetaGraphT *graph, TensorCache *tensor_cache,
-                         const QuantType &quant_type);
+                         schema::SubGraphT *sub_graph, schema::MetaGraphT *graph, const QuantType &quant_type);
 
-  STATUS ParseOnnxGivenFillNode(const onnx::NodeProto &onnx_node, TensorCache *tensor_cache);
+  STATUS ParseOnnxGivenFillNode(const onnx::NodeProto &onnx_node);
 
   STATUS ParseOnnxNodeAttr(const onnx::GraphProto &onnx_graph, const onnx::NodeProto &onnx_node,
                            const string &onnx_op_type, schema::CNodeT *dst_op);
 
   void SetOpQuantParams(const onnx::GraphProto &onnx_graph, const onnx::NodeProto &onnx_node, schema::CNodeT *dst_op,
-                        schema::TensorT *dst_tensor, TensorCache *tensor_cache);
+                        schema::TensorT *dst_tensor);
 
   STATUS SetOpInputIndex(const std::vector<string> &node_inputs, schema::CNodeT *dst_op,
-                         const onnx::NodeProto &onnx_node, TensorCache *tensor_cache);
+                         const onnx::NodeProto &onnx_node);
 
-  STATUS SetOpOutputIndex(const std::vector<string> &node_outputs, schema::CNodeT *dst_op, TensorCache *tensor_cache);
+  STATUS SetOpOutputIndex(const std::vector<string> &node_outputs, schema::CNodeT *dst_op);
 
   STATUS CopyOnnxTensorData(const onnx::TensorProto &onnx_init_value, schema::TensorT *tensor);
 
-  STATUS SetAllTensors(const TensorCache &tensor_cache, schema::MetaGraphT *graphDef);
+  STATUS SetAllTensors(schema::MetaGraphT *graphDef);
 
   void FindGraphInputAndConst(const onnx::GraphProto &onnx_graph);
 

mindspore/lite/tools/converter/parser/onnx/onnx_slice_parser.cc

@@ -17,9 +17,35 @@
 #include "tools/converter/parser/onnx/onnx_slice_parser.h"
 #include <memory>
 #include <vector>
+#include <string>
 
 namespace mindspore {
 namespace lite {
+STATUS OnnxSliceParser::InsertTensor(const std::vector<int> &onnx_val, const std::string &name,
+                                     onnx::NodeProto *onnx_node) {
+  std::unique_ptr<schema::TensorT> tensor = std::make_unique<schema::TensorT>();
+  if (tensor == nullptr) {
+    MS_LOG(ERROR) << "new tensor failed";
+    return RET_ERROR;
+  }
+  tensor->dataType = mindspore::kNumberTypeInt32;
+  tensor->dims.push_back(onnx_val.size());
+  tensor->format = schema::Format::Format_NCHW;
+  tensor->nodeType = schema::NodeType::NodeType_ValueNode;
+  int data_size = sizeof(int32_t) * onnx_val.size();
+  tensor->data.resize(data_size);
+  if (data_size != 0 &&
+      memcpy_s(static_cast<void *>(tensor->data.data()), data_size, onnx_val.data(), data_size) != EOK) {
+    MS_LOG(ERROR) << "memcpy_s failed";
+    return RET_ERROR;
+  }
+  int tensor_num = OnnxTensorParser::GetInstance()->GetTensorCache()->GetCachedTensor().size();
+  std::string tensor_name = name + std::to_string(tensor_num);
+  OnnxTensorParser::GetInstance()->GetTensorCache()->AddTensor(tensor_name, tensor.release(), GRAPH_INPUT);
+  onnx_node->add_input(tensor_name);
+  return RET_OK;
+}
+
 STATUS OnnxSliceParser::Parse(const onnx::GraphProto &onnx_graph, const onnx::NodeProto &onnx_node,
                               schema::CNodeT *op) {
   MS_LOG(DEBUG) << "onnx SliceParser";
@@ -33,15 +59,15 @@ STATUS OnnxSliceParser::Parse(const onnx::GraphProto &onnx_graph, const onnx::No
     return RET_NULL_PTR;
   }
 
-  std::unique_ptr<schema::SliceT> attr = std::make_unique<schema::SliceT>();
+  std::unique_ptr<schema::StridedSliceT> attr = std::make_unique<schema::StridedSliceT>();
   if (attr == nullptr) {
     MS_LOG(ERROR) << "new op failed";
     return RET_NULL_PTR;
   }
 
-  std::vector<int> axes;
   std::vector<int> starts;
   std::vector<int> ends;
+  std::vector<int> axes;
   std::vector<int> steps;
   for (const auto &onnx_node_attr : onnx_node.attribute()) {
     const auto &attribute_name = onnx_node_attr.name();
@@ -71,64 +97,49 @@ STATUS OnnxSliceParser::Parse(const onnx::GraphProto &onnx_graph, const onnx::No
       }
     }
   }
-
-  if (onnx_node.input_size() > 1) {
-    const auto &starts_name = onnx_node.input(1);
-    for (const auto &it : onnx_graph.initializer()) {
-      if (it.name() == starts_name) {
-        starts.clear();
-        for (int i = 0; i < it.int32_data_size(); ++i) {
-          starts.push_back(it.int32_data(i));
-        }
-      }
+  if (axes.empty()) {
+    for (size_t i = 0; i < starts.size(); ++i) {
+      axes.push_back(i);
     }
   }
-
-  if (onnx_node.input_size() > 2) {
-    const auto &ends_name = onnx_node.input(2);
-    for (const auto &it : onnx_graph.initializer()) {
-      if (it.name() == ends_name) {
-        ends.clear();
-        for (int i = 0; i < it.int32_data_size(); ++i) {
-          ends.push_back(it.int32_data(i));
-        }
-      }
-    }
+  if (steps.empty()) {
+    steps.assign(starts.size(), 1);
   }
-
-  if (onnx_node.input_size() > 3) {
-    const auto &axes_name = onnx_node.input(3);
-    for (const auto &it : onnx_graph.initializer()) {
-      if (it.name() == axes_name) {
-        axes.clear();
-        for (int i = 0; i < it.int32_data_size(); ++i) {
-          axes.push_back(it.int32_data(i));
-        }
-      }
+  onnx::NodeProto *slice_node = nullptr;
+  for (auto &node : onnx_graph.node()) {
+    if (&node == &onnx_node) {
+      slice_node = const_cast<onnx::NodeProto *>(&node);
     }
   }
-
-  if (onnx_node.input_size() > 4) {
-    const auto &steps_name = onnx_node.input(4);
-    for (const auto &it : onnx_graph.initializer()) {
-      if (it.name() == steps_name) {
-        steps.clear();
-        for (int i = 0; i < it.int32_data_size(); ++i) {
-          steps.push_back(it.int32_data(i));
-        }
-      }
+  int insert_num = 5 - onnx_node.input_size();
+  int status = RET_OK;
+  switch (insert_num) {
+    case 4: {
+      std::string name = "slice/starts/";
+      status = InsertTensor(starts, name, slice_node);
     }
+    case 3:
+      if (status == RET_OK) {
+        std::string name = "slice/ends/";
+        status = InsertTensor(ends, name, slice_node);
+      }
+    case 2:
+      if (status == RET_OK) {
+        std::string name = "slice/axes/";
+        status = InsertTensor(axes, name, slice_node);
+      }
+    case 1:
+      if (status == RET_OK) {
+        std::string name = "slice/steps/";
+        status = InsertTensor(steps, name, slice_node);
+      }
+    default:
+      if (status != RET_OK) {
+        MS_LOG(ERROR) << "onnx slice insert tensor failed";
+        return RET_ERROR;
+      }
   }
-
-  std::vector<int> sizes(starts.size(), -1);
-  for (size_t i = 0; i < starts.size(); ++i) {
-    sizes[i] = (ends[i] < 0 ? ends[i] : ends[i] - starts[i]);
-  }
-  attr->axes = axes;
-  attr->begin = starts;
-  attr->size = sizes;
-  attr->step = steps;
-  op->primitive->value.type = schema::PrimitiveType_Slice;
+  op->primitive->value.type = schema::PrimitiveType_StridedSlice;
   op->primitive->value.value = attr.release();
   return RET_OK;
 }

mindspore/lite/tools/converter/parser/onnx/onnx_slice_parser.h

@@ -17,8 +17,11 @@
 #ifndef MINDSPORE_LITE_TOOLS_CONVERTER_PARSER_ONNX_SLICE_PARSER_H
 #define MINDSPORE_LITE_TOOLS_CONVERTER_PARSER_ONNX_SLICE_PARSER_H
 
+#include <vector>
+#include <string>
 #include "tools/converter/parser/onnx/onnx_node_parser.h"
 #include "tools/converter/parser/onnx/onnx_node_parser_registry.h"
+#include "tools/converter/parser/onnx/onnx_tensor_parser.h"
 
 namespace mindspore {
 namespace lite {
@@ -27,6 +30,7 @@ class OnnxSliceParser : public OnnxNodeParser {
   OnnxSliceParser() : OnnxNodeParser("Slice") {}
   ~OnnxSliceParser() override = default;
 
+  STATUS InsertTensor(const std::vector<int> &onnx_val, const std::string &name, onnx::NodeProto *onnx_node);
   STATUS Parse(const onnx::GraphProto &onnx_graph, const onnx::NodeProto &onnx_node, schema::CNodeT *op) override;
 };
 }  // namespace lite

mindspore/lite/tools/converter/parser/onnx/onnx_tensor_parser.h

@@ -0,0 +1,39 @@
+/**
+ * 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_TOOLS_CONVERTER_PARSER_ONNX_TENSOR_PARSER_H
+#define MINDSPORE_LITE_TOOLS_CONVERTER_PARSER_ONNX_TENSOR_PARSER_H
+
+#include "tools/common/tensor_util.h"
+
+namespace mindspore {
+namespace lite {
+class OnnxTensorParser {
+ public:
+  ~OnnxTensorParser() = default;
+  static OnnxTensorParser *GetInstance() {
+    static OnnxTensorParser onnxTensorParser;
+    return &onnxTensorParser;
+  }
+  TensorCache *GetTensorCache() { return &tensor_cache_; }
+
+ private:
+  OnnxTensorParser() = default;
+  TensorCache tensor_cache_;
+};
+}  // namespace lite
+}  // namespace mindspore
+#endif  // MINDSPORE_LITE_TOOLS_CONVERTER_PARSER_ONNX_TESNOR_PARSER_H