Base class for detecting persons and feats from an image (cv::Mat).
Input is an image (cv::Mat).
Output is the enlarged image.
Sample code:
Note: The input image size is 640x480
auto image_file = string(argv[2]);
Mat input_img = imread(image_file);
if (input_img.empty()) {
cerr << "can't load image! " << argv[2] << endl;
return -1;
}
auto det = vitis::ai::FairMot::create(argv[1]);
auto result = det->run(input_img);
auto feats = result.feats;
auto bboxes = result.bboxes;
auto img = input_img.clone();
for (auto i = 0u; i < bboxes.size(); ++i) {
auto box = bboxes[i];
float x
= box.x * (img.cols);
float y
= box.y * (img.rows);
int xmin = x
;
int ymin = y
;
int xmax = x
+ (box.width) * (img.cols);
int ymax = y
+ (box.height) * (img.rows);
float score
= box.score;
xmin = std::min(std::max(xmin, 0), img.cols);
xmax = std::min(std::max(xmax, 0), img.cols);
ymin = std::min(std::max(ymin, 0), img.rows);
ymax = std::min(std::max(ymax, 0), img.rows);
LOG(INFO) << "RESULT " << box.label << " :\t" << xmin << "\t" << ymin
<< "\t" << xmax << "\t" << ymax << "\t" << score
<< "\n";
LOG(INFO) << "feat size: " << feats[i].size()
<< " First 5 digits: " << feats[i].data[0] + 0.0f << " "
<< feats[i].data[1] + 0.0f << " " << feats[i].data[2] + 0.0f
<< " " << feats[i].data[3] + 0.0f << " "
<< feats[i].data[4] + 0.0f << endl;
cv::rectangle(img, cv::Point(xmin, ymin), cv::Point(xmax, ymax),
cv::Scalar(0, 255, 0), 1, 1, 0);
}
auto out = image_file.substr(0, image_file.size() - 4) + "_out.jpg";
LOG(INFO) << "write result to " << out;
cv::imwrite(out, img);
Display of the model results:
Figure 1. result image

Struct of an object coordinates and confidence.
Declaration
typedef struct
{
float x,
float y,
float width,
float height,
int label,
float score
} vitis::ai::BoundingBox;
Member | Description |
---|---|
x |
x-coordinate. x is normalized relative to the input image columns. Range from 0 to 1. |
y |
y-coordinate. y is normalized relative to the input image rows. Range from 0 to 1. |
width |
Body width. Width is normalized relative to the input image columns, Range from 0 to 1. |
height |
Body height. Heigth is normalized relative to the input image rows, Range from 0 to 1. |
label | Body detection label. The value ranges from 0 to 21. |
score | Body detection confidence. The value ranges from 0 to 1. |