为什么Matlab给出的价值不同于C ++

时间:2013-03-26 16:23:41

标签: c++ matlab floating-point precision

我正在用c ++创建一个用于计算机视觉的庞大程序,为了调试目的,我使用了Matlab。经过一段时间的战斗,我意识到使用Matlab和c ++(Visual Studio 9编译器BTW)时,一些简单的算术方程的答案是不同的。那是为什么?

这是算术运算:

Matlab 功能:

function [x,y]=shape_fun(p,shape,Ax,Ay)

x=p(1)+Ax+shape(1)+Ax.*shape(3)+Ay.*shape(4)+Ax.*Ax.*shape(5)/2+Ay.*Ay.*shape(6)/2+Ax.*Ay.*shape(7);
y=p(2)+Ay+shape(2)+Ax.*shape(8)+Ay.*shape(9)+Ax.*Ax.*shape(10)/2+Ay.*Ay.*shape(11)/2+Ax.*Ay.*shape(12);

end

C ++ 功能

cv::Point2d deformed(const double shape_fun[12],const cv::Point2d p,const double Ax,const double Ay){
    cv::Point2d result;

    result.x=(p.x+Ax)+shape_fun[0]+shape_fun[2]*Ax+shape_fun[3]*Ay+shape_fun[4]*Ax*Ax/2*shape_fun[5]*Ay*Ay/2+shape_fun[6]*Ax*Ay;
    result.y=(p.y+Ay)+shape_fun[1]+shape_fun[7]*Ax+shape_fun[8]*Ay+shape_fun[9]*Ax*Ax/2*shape_fun[10]*Ay*Ay/2+shape_fun[11]*Ax*Ay;
    return result;

}

数据:

注意:Matlab值是从debbuger c ++值复制粘贴的,它们完全相同。

Ax=-12
Ay=-12
p=[468,683];
shape=[
     63.178114688537441
     36.536135487588474

    -0.038695673779030673
    -0.045313362559036965

     0.016469896824803026
     0.0017122284868442948
    -0.0030285669997117204

    -0.067902655024060773
     0.17995980761526389

     0.012716871878336870
    -0.036890386929202310
    -0.00081243692842574420
];

结果:

Matlab

x =
     3.947029931219995e+02
y =
     7.043339656551383e+02

C ++:

result  {x=393.54007007383439 y=703.64248713855773 }    cv::Point_<double>

注意:我没有故意将问题标记为OpenCV。虽然显然我在C ++中使用OpenCV,但cv::Point2d只是一个包含两个double变量x和y的结构,我不认为这个问题与OpenCV有关。一点都不。

1 个答案:

答案 0 :(得分:10)

result.x=
    (p.x+Ax)+
    shape_fun[0]+
    shape_fun[2]*Ax+
    shape_fun[3]*Ay+
    shape_fun[4]*Ax*Ax/2*
    shape_fun[5]*Ay*Ay/2+
    shape_fun[6]*Ax*Ay;
result.y=
    (p.y+Ay)+
    shape_fun[1]+
    shape_fun[7]*Ax+
    shape_fun[8]*Ay+
    shape_fun[9]*Ax*Ax/2*
    shape_fun[10]*Ay*Ay/2+
    shape_fun[11]*Ax*Ay;

应该(推测)

result.x=
    (p.x+Ax)+
    shape_fun[0]+
    shape_fun[2]*Ax+
    shape_fun[3]*Ay+
    shape_fun[4]*Ax*Ax/2+ /** change here **/
    shape_fun[5]*Ay*Ay/2+
    shape_fun[6]*Ax*Ay;
result.y=
    (p.y+Ay)+
    shape_fun[1]+
    shape_fun[7]*Ax+
    shape_fun[8]*Ay+
    shape_fun[9]*Ax*Ax/2+ /** change here **/
    shape_fun[10]*Ay*Ay/2+
    shape_fun[11]*Ax*Ay;