!5606 tflite custom parser

Merge pull request !5606 from 徐安越/master

build.sh

@@ -428,7 +428,7 @@ build_flatbuffer() {
     if [[ ! -f "${FLATC}" ]]; then
         git submodule update --init --recursive third_party/flatbuffers
         cd ${BASEPATH}/third_party/flatbuffers
-        rm -rf build && mkdir -pv build && cd build && cmake .. && make -j$THREAD_NUM
+        rm -rf build && mkdir -pv build && cd build && cmake -DFLATBUFFERS_BUILD_SHAREDLIB=ON .. && make -j$THREAD_NUM
         gene_flatbuffer
     fi
     if [[ "${INC_BUILD}" == "off" ]]; then

cmake/package_lite.cmake

@@ -67,6 +67,7 @@ else ()
     install(FILES ${TOP_DIR}/mindspore/lite/build/src/libmindspore-lite.so DESTINATION ${LIB_DIR_RUN_X86} COMPONENT ${RUN_X86_COMPONENT_NAME})
 
     install(FILES ${TOP_DIR}/third_party/protobuf/build/lib/libprotobuf.so.19.0.0 DESTINATION ${PROTOBF_DIR}/lib RENAME libprotobuf.so.19 COMPONENT ${COMPONENT_NAME})
+    install(FILES ${TOP_DIR}/third_party/flatbuffers/build/libflatbuffers.so.1.11.0 DESTINATION ${FLATBF_DIR}/lib RENAME libflatbuffers.so.1 COMPONENT ${COMPONENT_NAME})
 endif ()
 
 if (CMAKE_SYSTEM_NAME MATCHES "Windows")

mindspore/lite/schema/ops.fbs

@@ -355,6 +355,7 @@ table DetectionPostProcess {
     MaxClassesPreDetection: long;
     NumClasses: long;
     UseRegularNms: bool;
+    OutQuantized: bool;
 }
 
 table FullConnection {

mindspore/lite/src/ops/primitive_c.cc

@@ -120,6 +120,7 @@
 #include "src/ops/tuple_get_item.h"
 #include "src/ops/l2_norm.h"
 #include "src/ops/sparse_to_dense.h"
+#include "src/ops/detection_post_process.h"
 
 namespace mindspore {
 namespace lite {
@@ -469,6 +470,8 @@ PrimitiveC *PrimitiveC::UnPackFromSchemaPrimitiveT(mindspore::schema::PrimitiveT
       return new L2Norm(primitive);
     case schema::PrimitiveType_SparseToDense:
       return new SparseToDense(primitive);
+    case schema::PrimitiveType_DetectionPostProcess:
+      return new DetectionPostProcess(primitive);
     default:
       MS_LOG(ERROR) << "Unsupported primitive type in UnPackFromSchemaPrimitiveT : "
                     << schema::EnumNamePrimitiveType(op_type);
@@ -681,6 +684,8 @@ PrimitiveC *PrimitiveC::UnPackFromSchemaPrimitive(const schema::Primitive *primi
       return NewPrimitiveC<L2Norm>(primitive);
     case schema::PrimitiveType_SparseToDense:
       return NewPrimitiveC<SparseToDense>(primitive);
+    case schema::PrimitiveType_DetectionPostProcess:
+      return NewPrimitiveC<DetectionPostProcess>(primitive);
     default:
       MS_LOG(ERROR) << "Unsupported primitive type in UnPackFromSchemaPrimitive : "
                     << schema::EnumNamePrimitiveType(op_type);

mindspore/lite/test/run_benchmark_nets.sh

@@ -377,7 +377,7 @@ tar -zxf mindspore-lite-${version}-runtime-x86-${process_unit_x86}.tar.gz || exi
 tar -zxf mindspore-lite-${version}-converter-ubuntu.tar.gz || exit 1
 cd ${convertor_path}/mindspore-lite-${version}-converter-ubuntu || exit 1
 cp converter/converter_lite ./ || exit 1
-export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:./lib/:./third_party/protobuf/lib
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:./lib/:./third_party/protobuf/lib:./third_party/flatbuffers/lib
 
 # Convert the models
 cd ${convertor_path}/mindspore-lite-${version}-converter-ubuntu || exit 1

mindspore/lite/tools/anf_exporter/anf_exporter.cc

@@ -99,8 +99,7 @@ int AnfExporter::ConvertQuantParam(const std::unique_ptr<schema::MetaGraphT> &me
     auto output_quant_params = primitive->GetOutputQuantParams();
     if (output_quant_params.empty()) {
       if (node_type != schema::PrimitiveType_QuantDTypeCast) {
-        MS_LOG(ERROR) << "node: " << dst_node->name << " output quant params is empty";
-        return RET_ERROR;
+        MS_LOG(DEBUG) << "node: " << dst_node->name << " output quant params is empty";
       }
     } else {
       for (auto output_quant_param : output_quant_params[0]) {

mindspore/lite/tools/converter/legacy_optimizer/graph/dtype_trans_pass.cc

@@ -75,10 +75,9 @@ STATUS DTypeTransPass::DoModelInputDTypeTrans(schema::MetaGraphT *graph) {
     }
 
     for (auto iter = graph->nodes.begin(); iter != graph->nodes.end(); iter++) {
-      auto &node = *iter;
-      auto nodeName = node->name;
-      for (size_t inputIndexIdx = 0; inputIndexIdx < node->inputIndex.size(); inputIndexIdx++) {
-        if (node->inputIndex.at(inputIndexIdx) == graphInIdx) {
+      auto nodeName = (*iter)->name;
+      for (size_t inputIndexIdx = 0; inputIndexIdx < (*iter)->inputIndex.size(); inputIndexIdx++) {
+        if ((*iter)->inputIndex.at(inputIndexIdx) == graphInIdx) {
           STATUS status = RET_OK;
 
           // insert dtype cast node between input tensor and input node
@@ -108,11 +107,10 @@ STATUS DTypeTransPass::DoModelOutputDTypeTrans(schema::MetaGraphT *graph) {
   auto &graphOutIdxes = graph->outputIndex;
   for (auto graphOutIdx : graphOutIdxes) {
     for (auto iter = graph->nodes.begin(); iter != graph->nodes.end(); iter++) {
-      auto &node = *iter;
-      auto nodeName = node->name;
+      auto nodeName = (*iter)->name;
       MS_ASSERT(node != nullptr);
-      for (size_t outputIndexIdx = 0; outputIndexIdx < node->outputIndex.size(); outputIndexIdx++) {
-        if (node->outputIndex.at(outputIndexIdx) == graphOutIdx) {
+      for (size_t outputIndexIdx = 0; outputIndexIdx < (*iter)->outputIndex.size(); outputIndexIdx++) {
+        if ((*iter)->outputIndex.at(outputIndexIdx) == graphOutIdx) {
           // insert transNode
           STATUS status = RET_OK;
           iter = InsertDTypeTransNode(graph, iter, kAfter, outputIndexIdx, kInt8ToFP32, &status);
@@ -135,7 +133,6 @@ STATUS DTypeTransPass::DoNodeInoutDTypeTrans(schema::MetaGraphT *graph) {
     if (IsContain(GetUint8OpList(), GetCNodeTType(**iter)) && (*iter)->quantType == QuantType_AwareTraining) {
       continue;
     }
-    auto &node = *iter;
     if (GetCNodeTType(**iter) == PrimitiveType_QuantDTypeCast) {
       continue;
     }
@@ -143,8 +140,8 @@ STATUS DTypeTransPass::DoNodeInoutDTypeTrans(schema::MetaGraphT *graph) {
     if (GetCNodeTType(**iter) == PrimitiveType_Shape) {
       needInsertPost = false;
     }
-    auto nodeName = node->name;
-    if (node->inputIndex.size() < kMinInputNum) {
+    auto nodeName = (*iter)->name;
+    if ((*iter)->inputIndex.size() < kMinInputNum) {
       MS_LOG(ERROR) << "Op " << nodeName.c_str() << " should have " << kMinInputNum << " input tensor at least";
       return RET_ERROR;
     }

mindspore/lite/tools/converter/parser/tflite/CMakeLists.txt

@@ -1,6 +1,9 @@
 file(GLOB_RECURSE TFLITE_SRC_LIST RELATIVE ${CMAKE_CURRENT_SOURCE_DIR}
         *.cc
         )
+ADD_DEFINITIONS(-DFLATBUFFERS_LOCALE_INDEPENDENT=1)
+find_library(FLATBUFFERS_LIBRARY flatbuffers HINTS ${TOP_DIR}/third_party/flatbuffers/build)
 add_library(tflite_parser_mid OBJECT
         ${TFLITE_SRC_LIST}
         )
+target_link_libraries(tflite_parser_mid ${FLATBUFFERS_LIBRARY})

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

@@ -18,6 +18,8 @@
 #include <vector>
 #include <memory>
 #include <map>
+#include "flatbuffers/flatbuffers.h"
+#include "flatbuffers/flexbuffers.h"
 
 namespace mindspore {
 namespace lite {
@@ -39,15 +41,42 @@ STATUS TfliteCustomParser::Parse(const std::unique_ptr<tflite::OperatorT> &tflit
     return RET_NULL_PTR;
   }
 
-  std::unique_ptr<schema::CustomT> attr = std::make_unique<schema::CustomT>();
+  std::unique_ptr<schema::DetectionPostProcessT> attr = std::make_unique<schema::DetectionPostProcessT>();
   if (attr == nullptr) {
     MS_LOG(ERROR) << "new op failed";
     return RET_NULL_PTR;
   }
 
   const auto &custom_attr = tflite_op->custom_options;
-  attr->custom = custom_attr;
-  op->primitive->value.type = schema::PrimitiveType_Custom;
+  auto attr_map = flexbuffers::GetRoot(custom_attr).AsMap();
+  attr->format = schema::Format_NHWC;
+  attr->inputSize = tflite_op->inputs.size();
+  attr->hScale = attr_map["h_scale"].AsFloat();
+  attr->wScale = attr_map["w_scale"].AsFloat();
+  attr->xScale = attr_map["x_scale"].AsFloat();
+  attr->yScale = attr_map["y_scale"].AsFloat();
+  attr->NmsIouThreshold = attr_map["nms_iou_threshold"].AsFloat();
+  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;
+  } else {
+    attr->DetectionsPreClass = attr_map["detections_per_class"].AsInt32();
+  }
+  attr->MaxClassesPreDetection = attr_map["max_classes_per_detection"].AsInt32();
+  attr->NumClasses = attr_map["num_classes"].AsInt32();
+  if (attr_map["use_regular_nms"].IsNull()) {
+    attr->UseRegularNms = false;
+  } else {
+    attr->UseRegularNms = attr_map["use_regular_nms"].AsBool();
+  }
+  if (attr_map["_output_quantized"].IsNull()) {
+    attr->OutQuantized = false;
+  } else {
+    attr->OutQuantized = attr_map["_output_quantized"].AsBool();
+  }
+
+  op->primitive->value.type = schema::PrimitiveType_DetectionPostProcess;
   op->primitive->value.value = attr.release();
 
   for (size_t i = 0; i < tflite_op->inputs.size(); ++i) {