saved_model_utils.h

Go to the documentation of this file.

/*
==============================================================================
MIT License
Copyright 2022 Institute for Automotive Engineering of RWTH Aachen University.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
==============================================================================
*/
 
#pragma once
 
#include <algorithm>
#include <sstream>
#include <stdexcept>
#include <string>
#include <vector>
 
#include <tensorflow/cc/saved_model/loader.h>
#include <tensorflow/cc/saved_model/tag_constants.h>
#include <tensorflow_cpp/utils.h>
 
 
namespace tensorflow_cpp {
 
 
namespace tf = tensorflow;
 
 
inline tf::SavedModelBundleLite loadSavedModel(
  const std::string& dir, const bool allow_growth = true,
  const double per_process_gpu_memory_fraction = 0,
  const std::string& visible_device_list = "") {
 
  tf::SavedModelBundleLite saved_model;
  tf::SessionOptions session_options = makeSessionOptions(
    allow_growth, per_process_gpu_memory_fraction, visible_device_list);
  tf::Status status =
    tf::LoadSavedModel(session_options, tf::RunOptions(), dir,
                       {tf::kSavedModelTagServe}, &saved_model);
  if (!status.ok())
    throw std::runtime_error("Failed to load SavedModel: " + status.ToString());
 
  return saved_model;
}
 
 
inline tf::Session* loadSavedModelIntoNewSession(
  const std::string& dir, const bool allow_growth = true,
  const double per_process_gpu_memory_fraction = 0,
  const std::string& visible_device_list = "") {
 
  tf::SavedModelBundleLite saved_model = loadSavedModel(
    dir, allow_growth, per_process_gpu_memory_fraction, visible_device_list);
  tf::Session* session = saved_model.GetSession();
 
  return session;
}
 
 
inline tf::Session* getSessionFromSavedModel(
  const tf::SavedModelBundleLite& saved_model) {
 
  return saved_model.GetSession();
}
 
 
inline std::string getSavedModelNodeByLayerName(
  const tf::SavedModelBundleLite& saved_model, const std::string& layer_name,
  const std::string& signature = "serving_default") {
 
  std::string node_name;
  const tf::SignatureDef& model_def = saved_model.GetSignatures().at(signature);
  auto inputs = model_def.inputs();
  auto outputs = model_def.outputs();
  auto& nodes = inputs;
  nodes.insert(outputs.begin(), outputs.end());
  for (const auto& node : nodes) {
    const std::string& key = node.first;
    const tf::TensorInfo& info = node.second;
    if (key == layer_name) {
      node_name = info.name();
      break;
    }
  }
 
  return node_name;
}
 
 
inline std::string getSavedModelLayerByNodeName(
  const tf::SavedModelBundleLite& saved_model, const std::string& node_name,
  const std::string& signature = "serving_default") {
 
  std::string layer_name;
  const tf::SignatureDef& model_def = saved_model.GetSignatures().at(signature);
  auto inputs = model_def.inputs();
  auto outputs = model_def.outputs();
  auto& nodes = inputs;
  nodes.insert(outputs.begin(), outputs.end());
  for (const auto& node : nodes) {
    const std::string& key = node.first;
    const tf::TensorInfo& info = node.second;
    if (info.name() == node_name) {
      layer_name = key;
      break;
    }
  }
 
  return layer_name;
}
 
 
inline std::vector<std::string> getSavedModelInputNames(
  const tf::SavedModelBundleLite& saved_model, const bool layer_names = false,
  const std::string& signature = "serving_default") {
 
  std::vector<std::string> names;
  const tf::SignatureDef& model_def = saved_model.GetSignatures().at(signature);
  for (const auto& node : model_def.inputs()) {
    const std::string& key = node.first;
    const tf::TensorInfo& info = node.second;
    names.push_back(info.name());
  }
  std::sort(names.begin(), names.end());
 
  if (layer_names) {
    std::vector<std::string> node_names = names;
    names = {};
    for (const auto& node_name : node_names)
      names.push_back(
        getSavedModelLayerByNodeName(saved_model, node_name, signature));
  }
 
  return names;
}
 
 
inline std::vector<std::string> getSavedModelOutputNames(
  const tf::SavedModelBundleLite& saved_model, const bool layer_names = false,
  const std::string& signature = "serving_default") {
 
  std::vector<std::string> names;
  const tf::SignatureDef& model_def = saved_model.GetSignatures().at(signature);
  for (const auto& node : model_def.outputs()) {
    const std::string& key = node.first;
    const tf::TensorInfo& info = node.second;
    names.push_back(info.name());
  }
  std::sort(names.begin(), names.end());
 
  if (layer_names) {
    std::vector<std::string> node_names = names;
    names = {};
    for (const auto& node_name : node_names)
      names.push_back(
        getSavedModelLayerByNodeName(saved_model, node_name, signature));
  }
 
  return names;
}
 
 
inline std::vector<int> getSavedModelNodeShape(
  const tf::SavedModelBundleLite& saved_model, const std::string& node_name,
  const std::string& signature = "serving_default") {
 
  std::vector<int> node_shape;
  const tf::SignatureDef& model_def = saved_model.GetSignatures().at(signature);
  auto inputs = model_def.inputs();
  auto outputs = model_def.outputs();
  auto& nodes = inputs;
  nodes.insert(outputs.begin(), outputs.end());
  for (const auto& node : nodes) {
    const std::string& key = node.first;
    const tf::TensorInfo& info = node.second;
    if (info.name() == node_name) {
      const auto& shape = info.tensor_shape();
      for (int d = 0; d < shape.dim_size(); d++)
        node_shape.push_back(shape.dim(d).size());
      break;
    }
  }
 
  return node_shape;
}
 
 
inline tf::DataType getSavedModelNodeType(
  const tf::SavedModelBundleLite& saved_model, const std::string& node_name,
  const std::string& signature = "serving_default") {
 
  tf::DataType type = tf::DT_INVALID;
  const tf::SignatureDef& model_def = saved_model.GetSignatures().at(signature);
  auto inputs = model_def.inputs();
  auto outputs = model_def.outputs();
  auto& nodes = inputs;
  nodes.insert(outputs.begin(), outputs.end());
  for (const auto& node : nodes) {
    const std::string& key = node.first;
    const tf::TensorInfo& info = node.second;
    if (info.name() == node_name) {
      type = info.dtype();
      break;
    }
  }
 
  return type;
}
 
 
inline std::string getSavedModelInfoString(
  const tf::SavedModelBundleLite& saved_model) {
 
  std::stringstream ss;
  ss << "SavedModel Info:" << std::endl;
 
  ss << "Signatures:" << std::endl;
  const auto& signatures = saved_model.GetSignatures();
  for (const auto& sig : signatures) {
 
    ss << "  " << sig.first << std::endl;
    const auto& def = sig.second;
 
    ss << "    Inputs: " << def.inputs_size() << std::endl;
    for (const auto& node : def.inputs()) {
      ss << "      " << node.first << ": " << node.second.name() << std::endl;
      ss << "        Shape: [ ";
      for (int d = 0; d < node.second.tensor_shape().dim_size(); d++) {
        ss << node.second.tensor_shape().dim(d).size() << ", ";
      }
      ss << "]" << std::endl;
      ss << "        DataType: " << tf::DataTypeString(node.second.dtype())
         << std::endl;
    }
 
    ss << "    Outputs: " << def.outputs_size() << std::endl;
    for (const auto& node : def.outputs()) {
      ss << "      " << node.first << ": " << node.second.name() << std::endl;
      ss << "        Shape: [ ";
      for (int d = 0; d < node.second.tensor_shape().dim_size(); d++) {
        ss << node.second.tensor_shape().dim(d).size() << ", ";
      }
      ss << "]" << std::endl;
      ss << "        DataType: " << tf::DataTypeString(node.second.dtype())
         << std::endl;
    }
  }
 
  return ss.str();
}
 
 
}  // namespace tensorflow_cpp