54 lines
1.8 KiB
C++
54 lines
1.8 KiB
C++
//
|
|
// Created by DefTruth on 2021/6/14.
|
|
//
|
|
|
|
#include "mobilenetv2.h"
|
|
#include "lite/ort/core/ort_utils.h"
|
|
#include "lite/utils.h"
|
|
|
|
using ortcv::MobileNetV2;
|
|
|
|
Ort::Value MobileNetV2::transform(const cv::Mat &mat)
|
|
{
|
|
cv::Mat canvas;
|
|
cv::resize(mat, canvas, cv::Size(input_node_dims.at(3),
|
|
input_node_dims.at(2)));
|
|
cv::cvtColor(canvas, canvas, cv::COLOR_BGR2RGB);
|
|
|
|
canvas.convertTo(canvas, CV_32FC3, 1. / 255.f, 0.f);
|
|
ortcv::utils::transform::normalize_inplace(canvas, mean_vals, scale_vals); // float32
|
|
// (1,3,224,224)
|
|
return ortcv::utils::transform::create_tensor(
|
|
canvas, input_node_dims, memory_info_handler,
|
|
input_values_handler, ortcv::utils::transform::CHW);
|
|
}
|
|
|
|
void MobileNetV2::detect(const cv::Mat &mat, types::ImageNetContent &content, unsigned int top_k)
|
|
{
|
|
if (mat.empty()) return;
|
|
// 1. make input tensor
|
|
Ort::Value input_tensor = this->transform(mat);
|
|
// 2. inference
|
|
auto output_tensors = ort_session->Run(
|
|
Ort::RunOptions{nullptr}, input_node_names.data(),
|
|
&input_tensor, 1, output_node_names.data(), num_outputs
|
|
);
|
|
Ort::Value &logits_tensor = output_tensors.at(0); // (1,1000)
|
|
const unsigned int num_classes = output_node_dims.at(0).at(1);
|
|
const float *logits = logits_tensor.GetTensorMutableData<float>(); // float
|
|
unsigned int max_id;
|
|
std::vector<float> scores = lite::utils::math::softmax<float>(logits, num_classes, max_id);
|
|
std::vector<unsigned int> sorted_indices = lite::utils::math::argsort<float>(scores);
|
|
if (top_k > num_classes) top_k = num_classes;
|
|
|
|
content.scores.clear();
|
|
content.labels.clear();
|
|
content.texts.clear();
|
|
for (unsigned int i = 0; i < top_k; ++i)
|
|
{
|
|
content.labels.push_back(sorted_indices[i]);
|
|
content.scores.push_back(scores[sorted_indices[i]]);
|
|
content.texts.push_back(class_names[sorted_indices[i]]);
|
|
}
|
|
content.flag = true;
|
|
} |