python numpy to CSV, numpy to pandas, pandas to CSV

Sample code for
- numpy to CSV
- numpy to pandas
- pandas to CSV

import numpy as np
import pandas as pd

f1_numpy = "./data/test1.csv"
f2_pandas = "./data/test2.csv"

#numpy to csv
np.savetxt(f1_numpy, np.array([10,20]))
print(f1_numpy)

#numpy to pandas and csv
pda = pd.DataFrame(np.array([10,20]), columns=['data'])
pda.to_csv(f2_pandas, index=False)

Thank you.

opencv perspective test

This is perspective example source code.
Basically, perspective is multiple by H(homography) matrix.
Like this :
B = H*A
In here, B is perspective image, A is original image and H is homography matrix.
For this calculate, we can use opencv function this -> warpPerspective(..)
 easily.
And to find H, we also can use findHomography opencv function as well.

for more detail refer to this page:
http://study.marearts.com/2014/04/video-stabilizing-example-source-code.html
http://study.marearts.com/2011/10/sift-matching-c-source-code-using.html
http://study.marearts.com/2015/03/image-warping-using-opencv.html

and below source code:
code looks a little complicated, but just see getPerspectiveImg function carefully, other codes are just for making random and set value.

github url is here : https://github.com/MareArts/opencv_perspective_test

Thank you.

#include <stdio.h>
#include <opencv2/opencv.hpp> 
#include <string>

#ifdef _DEBUG  
#pragma comment(lib, "opencv_core331d.lib")   
#pragma comment(lib, "opencv_imgproc331d.lib")   //MAT processing  
#pragma comment(lib, "opencv_highgui331d.lib")  
#pragma comment(lib, "opencv_calib3d331d.lib") 
#pragma comment(lib, "opencv_imgcodecs331d.lib")
#else  
#pragma comment(lib, "opencv_core331.lib")  
#pragma comment(lib, "opencv_imgproc331.lib")  
#pragma comment(lib, "opencv_highgui331.lib")
#pragma comment(lib, "opencv_calib3d331.lib")
#pragma comment(lib, "opencv_imgcodecs331.lib")
#endif


//#pragma comment(lib, "opencv_objdetect246.lib")  
//#pragma comment(lib, "opencv_features2d246.lib")  


using namespace std;
using namespace cv;


bool setPt(float x1, float x2, float x3, float x4, float y1, float y2, float y3, float y4, vector< Point2f> &in)
{
 if (in.size() == 4)
 {
  in[0].x = x1;
  in[1].x = x2;
  in[2].x = x3;
  in[3].x = x4;

  in[0].y = y1;
  in[1].y = y2;
  in[2].y = y3;
  in[3].y = y4;
 }
 else {
  return false;
 }

 return true;
}


bool makeRandomPerspectivePoint(const vector< Point2f> &A, vector< Point2f> &B, Size imgSize)
{
 if (A.size() != 4 || B.size() != 4)
 {
  cout << "matrix vector error!" << endl;
  return false;
 }

 if (imgSize.width <= 0 || imgSize.height <= 0)
 {
  cout << "image size error!" << endl;
  return false;
 }

 const float maxRand = 0.4;
 int width = imgSize.width;
 int height = imgSize.height;
 int rndMarginX = width * maxRand;
 int rndMarginY = height * maxRand;

 int rx1 = rand() % rndMarginX;
 int ry1 = rand() % rndMarginY;
 int rM = rand() % 4;

 //minimum value
 if (rx1 == 0)
  rx1 = 1;
 if (ry1 == 0)
  ry1 = 1;

 //4 perspective mode
 if (rM == 0)
 {
  setPt(0, width, width, 0, 0 + ry1, 0, height, height - ry1, B);
 }
 else if (rM == 1)
 {
  setPt(0 + rx1, width - rx1, width, 0, 0, 0, height, height, B);
 }
 else if (rM == 2)
 {
  setPt(0, width, width, 0, 0, 0 + ry1, height - ry1, height, B);
 }
 else {
  setPt(0, width, width - rx1, 0 + rx1, 0, 0, height, height, B);
 }

 return true;
}

bool getPerspectiveImg(const Mat& inImg, Mat& outImg, const vector< Point2f> &A, const vector< Point2f> &B)
{
 if (A.size() != 4 || B.size() != 4)
 {
  cout << "matrix vector error!" << endl;
  return false;
 }

 //B = H*A
 Mat H = findHomography(A, B);
 //perspectiveImg = H * inImg
 warpPerspective(inImg, outImg, H, inImg.size());


 //return
 return true;
}



int main()
{
 
 string fn = "Flag_of_South_Korea.png";
 Mat testImg = imread(fn);
 srand(time(0));
 Size imgSize = testImg.size();
 vector< Point2f > opt(4);
 setPt(0, imgSize.width, imgSize.width, 0, 0, 0, imgSize.height, imgSize.height, opt);
 vector< Point2f > dpt(4);

 while (1)
 {

  if (makeRandomPerspectivePoint(opt, dpt, imgSize) == false)
   break;

  Mat outImg;
  if (getPerspectiveImg(testImg, outImg, opt, dpt)) {
   imshow("origin", testImg);
   imshow("perspective", outImg);
   waitKey(0);
  }
  else {
   cout << "matrix vector error!" << endl;
   break;
  }
 }

}

3D array numpy -> pandas ->csv -> pandas -> 3d array numpy

This article is example source code for
3D array numpy -> pandas -> csv -> pandas -> 3D array numpy

Let's see step by step


Step 1, make example data

import numpy as np
import pandas as pd


#make list
a = [[11, 12, 13, 14, 15], [15, 16, 17, 18, 19]]
b = [[21, 22, 23, 24, 25], [25, 26, 27, 28, 29]]
c = []
c.append(a)
c.append(b)
#make numpy
npa = np.array(c)
print('npa\n',npa)
print('npa shape\n',npa.shape) #2 by 2 by 5

result

npa
 [[[11 12 13 14 15]
  [15 16 17 18 19]]

 [[21 22 23 24 25]
  [25 26 27 28 29]]]
npa shape
 (2, 2, 5)

Step 2, numpy to pandas

#make numpy to panda
m,n,r = npa.shape
#numpy ->group indexing, reshape
out_arr = np.column_stack((np.repeat(np.arange(m),n),npa.reshape(m*n,-1)))
out_df = pd.DataFrame(out_arr, columns=['group','a','b','c','d','e'])
print('pnadas\n',out_df) #pandas

result

group   a   b   c   d   e
0      0  11  12  13  14  15
1      0  15  16  17  18  19
2      1  21  22  23  24  25
3      1  25  26  27  28  29

Step 3, save csv, load csv

#save to csv
out_df.to_csv('test3Dpandas.csv', index=False)
#load csv
df = pd.read_csv('test3Dpandas.csv')

Step 4, pandas to numpy

#pandas to numpy
npb = df.values
npb = npb[:,1:]
npb2 = npb.reshape(m,n,r)
print('numpy\n',npb2)

result

numpy
 [[[11 12 13 14 15]
  [15 16 17 18 19]]

 [[21 22 23 24 25]
  [25 26 27 28 29]]]