Trying to use TensorFlow C++ API with Cast operation (Movenet Model) results in error when inference

Question

Hi I'm trying to use the new MoveNet Model with the tensorlow C++ API.

This is my part of the code trying to load the SavedModel and the image.

Load Model

// Load
SavedModelBundleLite  model_bundle;
SessionOptions session_options = SessionOptions();
session_options.config.mutable_gpu_options()->set_allow_growth(true);
RunOptions run_options = RunOptions();
std::string export_dir = "/workspace/modelZoo";
Status status = LoadSavedModel(session_options, run_options, export_dir, {"serve"}, &model_bundle);

Load Image

//Load image
int32_t input_width = 256;
int32_t input_height = 256;

using namespace ::tensorflow::ops;
std::string filename = "image.jpg";
Scope root = Scope::NewRootScope();
auto output = tensorflow::ops::ReadFile(root.WithOpName("file_reader"), filename);

const int wanted_channels = 3;
tensorflow::Output image_reader = tensorflow::ops::DecodeJpeg(root.WithOpName("file_decoder"), output, tensorflow::ops::DecodeJpeg::Channels(wanted_channels));

auto image_int32 = tensorflow::ops::Cast(root.WithOpName("int32_caster"), image_reader, tensorflow::DT_INT32);
auto dims_expander = tensorflow::ops::ExpandDims(root.WithOpName("expand_dims"), image_int32, 0);
auto resized = tensorflow::ops::ResizeBilinear(root, dims_expander, tensorflow::ops::Const(root.WithOpName("size"), {input_height, input_width}));

std::vector<Tensor> out_tensors;
ClientSession session(root);
auto run_status = session.Run({resized}, &out_tensors);

And INFERENCE

const std::string input_name_1 = model_def.inputs().at("input").name();
std::vector<std::pair<string, Tensor>> inputs_data  = {{input_name_1, out_tensors[0]}};


std::string outputLayer = model_def.outputs().at("output_0").name();
std::vector<Tensor> outputs;

Status runStatus = model_bundle.GetSession()->Run(inputs_data, {outputLayer}, {}, &outputs);

When the code runs the inference I get this error.

Expects arg[0] to be int32 but float is provided

Based on the tutorial from tensorflow hub (https://tfhub.dev/google/movenet/multipose/lightning/1) the model needs int32 format but I don't know what is wrong with my code referring to the casting part of the input tensor