Develop Reference

What is the best way to process output of segmentation network in Microsoft.ML?

The network produces 1 x N x K tensor, where N is number of pixel positions and K is number of classes, each value represents score for a class at given position.
Current code to retrieve best class affinity for each position is working, but it is terribly slow and takes x4 more time, than the network run itself.
private int[,] GetClasses(List<DisposableNamedOnnxValue> output)
{
Tensor<float> outTensor = output.First().AsTensor<float>();
int[,] classes = new int[frameWidth,frameHeight];
for (int i = 0; i < frameWidth; ++i)
{
for (int j = 0; j < frameHeight; ++j)
{
int finalClass = 0;
float finalClassScore = 0;
for (int k = 0; k < nClasses; ++k)
{
float score = outTensor[0, i * frameHeight + j, k];
if (score > finalClassScore)
{
finalClassScore = score;
finalClass = k;
}
}
classes[i, j] = finalClass;
}
}
return classes;
}
Is there a better, faster way of doing this in Microsoft.ML ?

The solution I went with was to add argmax layer to the initial keras model. Keras output single value through argmax.

multidimenional class arrays from multidimensional array c#

I have a 9x9 array of values. I would like to divide those values up into 3x3 "Cells". These "Cells" are class objects that are also basically a 3x3 array. My problem is, when creating the loop, something doesn't happen correctly. I've added a line for debugging that puts out "Table[0,0].Item(0,0)", where "Table" is the 3x3 array of class objects and ".Item" is a function that returns what is at index .Item(x,y) in the "Cell". Unfortunately, for some reason this line for debugging puts out different values every time it loops. I have just gotten into coding, and I'm sure there's tons of better ways to solve this, but first I would like to figure out why my current code is not working to learn.
This is where I try and load the Table: ("table" with a lower case in the original 9x9 array)
static void Load_Table()
{
for (int i = 0; i < 3; i++)
{
for (int j = 0; j < 3; j++)
{
Console.WriteLine("Current cell: " + i + "/" + j);
Cell tempCell = new Cell(table, i, j);
Table[i, j] = tempCell;
Console.WriteLine(Table[0, 0].Item(0, 0));
}
}
}
The Cell class:
class Cell
{
public static int[] Cellvalue = new int[2];
public static int[,] content = new int[3, 3];
private int[] contains = new int[9];
public Cell(int[,] table,int column, int row)
{
Cellvalue[0] = column; Cellvalue[1] = row;
for (int i = 0;i < 3; i++)
{
for (int j = 0; j < 3; j++)
{
//Console.WriteLine("{0}{1} {2}{3}",Cellvalue[0],Cellvalue[1],i,j);
content[i, j] = table[i + (Cellvalue[0] * 3), j+ (Cellvalue[1] * 3)];
}
}
}
public int Item(int column,int row)
{
return content[column, row];
}
}
And table and Table:
static public int[,] table = new int[9, 9];
static public Cell[,] Table = new Cell[3, 3];
Thank you very much for the help in advance kind strangers <3.

Show sublist value from index C#

My problem is I have M subproblem. Each subproblem include x (double array) and reduced cost which has different value for each iteration m. I want to show x which has the minimum reduced cost among all subproblems. here is my class:
public class Subproblem
{
public double[,] x { get; set; }
public double ReducedCost { get; set; }
}
So far, I already can get the minimum Reduced Cost and index of it. Now I want to show x value (double array) which on that index. I've code like this:
var sub = new List<Subproblem>();
for (int m = 0; m < M; ++m)
{
Subproblem s = new Subproblem();
s.x = new double[DC1, DC1];
s.ReducedCost = model.ObjVal;
for (int i = 0; i < DC1; ++i)
{
for (int j = 0; j < DC1; ++j)
{
s.x[i, j] = x[i, j].X;
}
}
sub.Add(s);
}
double minRC = sub.Min(a => a.ReducedCost);
int minRCIndex = sub.FindIndex((i) => i.ReducedCost == minRC);
Console.WriteLine(sub.x(minRCIndex));
the last line (Console.WriteLine(sub.x(minRCIndex));) still got red underline, I don't know how to write it

If you are only trying to get the minimum Reduced Costed subproblem you should be doing:
Subproblem minimumReducedCostedSubproblem = sub[minRCIndex];
And you can print the matrix down like this:
for (int i = 0; i < DC1; ++i)
{
for (int j = 0; j < DC1; ++j)
{
Console.Write(minimumReducedCostedSubproblem.x[i, j] + "\t");
}
Console.Write("\n");
}
But you seem a little confused. You are pushing a Subproblem into your sub list with the same object for M times. Because model.ObjVal doesn't change along the first for loop. There is something wierd going on there too.

It should be
var objWithMinReduceCost = sub[minRCIndex];
//Now you have the object of Subproblem derived from your logic.
//You can access x property of it have further logic to process it.
for (int i = 0; i < DC1; ++i)
{
for (int j = 0; j < DC1; ++j)
{
Console.WriteLine(objWithMinReduceCost.x[i, j]);
}
}

If you are interested in obtaining the double array, you could just do this:
double[,] result = sub.First(i => i.ReducedCost == minRC).x;
Although as Tolga mentioned, all your elements will have the same ReducedCost with your current code.

What is the most efficient way to find the largest element in a matrix along with position? Also, the largest element in each column with position

I have written the following code but it looks to be far from efficient.
//Find largest in tempRankingData
int largestIntempRankingData = tempRankingData[0, 0];
for (int i = 0; i < count; i++)
{
for (int j = 0; j < count; j++)
{
if (tempRankingData[i, j] > largestIntempRankingData)
{
largestIntempRankingData = tempRankingData[i, j];
}
}
}
//Find position of largest in tempRankingData
List<string> positionLargestIntempRankingData = new List<string>();
for (int i = 0; i < count; i++)
{
for (int j = 0; j < count; j++)
{
if (tempRankingData[i, j] == largestIntempRankingData)
{
positionLargestIntempRankingData.Add(i + "," + j);
}
}
}
//Find largest in each column
int largestInColumn = 0;
List<string> positionOfLargestInColumn = new List<string>();
Dictionary<int, List<string>> position = new Dictionary<int, List<string>>();
for (int i = 0; i < count; i++)
{
largestInColumn = tempRankingData[0, i];
positionOfLargestInColumn = new List<string>();
for (int j = 0; j < count; j++)
{
if (tempRankingData[j, i] > largestInColumn)
{
largestInColumn = tempRankingData[j, i];
}
}
for (int j = 0; j < count; j++)
{
if (tempRankingData[j, i] == largestInColumn)
{
positionOfLargestInColumn.Add(j + "," + i);
}
}
position.Add(i, positionOfLargestInColumn);
}
So, I wanted to check about the most efficient way to do this.

Whilst you're finding the largest in each column, you could also be finding the largest overall. You can also capture the positions as you go:
//Find largest in each column
int largestInColumn = 0;
int largestOverall = int.MinValue;
List<string> positionOfLargestInColumn;
Dictionary<int, List<string>> position = new Dictionary<int, List<string>>();
List<string> positionLargestIntempRankingData = new List<string>();
for (int i = 0; i < count; i++)
{
largestInColumn = tempRankingData[0, i];
positionOfLargestInColumn = new List<string>();
positionOfLargestInColumn.Add("0," + i);
for (int j = 1; j < count; j++)
{
if (tempRankingData[j, i] > largestInColumn)
{
largestInColumn = tempRankingData[j, i];
positionOfLargestInColumn.Clear();
positionOfLargestInColumn.Add(j + "," + i);
}
else if(tempTankingData[j,i] == largestInColumn)
{
positionOfLargestInColumn.Add(j + "," + i);
}
}
position.Add(i, positionOfLargestInColumn);
if(largestInColumn > largestOverall)
{
positionLargestIntempRankingData.Clear();
positionLargestIntempRankingData.AddRange(positionOfLargestInColumn);
largestOverall = largestInColumn;
}
else if(largestInColumn == largestOverall)
{
positionLargestIntempRankingData.AddRange(positionOfLargestInColumn);
}
}

1). You can find largest element and its position in one method and retrieve.
Would be caller of your method concerned about position or actual value, is a matter of concrete case.
2) You can use `yield return' technique in your matrix search (for column based search), so do not compute all column's maximas and push them into the dictionary. Dictionaries are not that fast as arrays, if you can avoid use them, do that.
3) You can keep a matrix in single dimension, long array. Have [] access operator overload, to "emulate" matrix access. Why ? If finding maximum is something frequent you might need to do during program run, having one foreach loop is faster then having 2 nested once. In case of a big matrices, single array search can be easily parallelized among different cores.
If big matrices and/or frequent calls are not your concern, just simplify your code like in points (1), (2).

For your fist two itterations you could replace with this:
//Find largest in tempRankingData
int largestIntempRankingData = tempRankingData[0, 0];
List<KeyValuePair<double,string>> list = new List<KeyValuePair<double,string>>();
for (int i = 0; i < count; i++)
{
for (int j = 0; j < count; j++)
{
if (tempRankingData[i, j] > largestIntempRankingData)
{
largestIntempRankingData = tempRankingData[i, j];
list.Add(new KeyValuePair<double, string>(largestIntempRankingData, i + "," + j)); //Add the value and the position;
}
}
}
//This gives a list of strings in which hold the position of largestInItemRankingData example "3,3"
//Only positions where the key is equal to the largestIntempRankingData;
list.Where(w => w.Key == largestIntempRankingData).ToList().Select(s => s.Value).ToList();

You can get all these pieces of information in a single scan with a little fiddling around. Something like this (converting the rows and columns to a string is trivial and better done at the end anyway):
int? largestSoFar = null; // you could populate this with myMatrix[0,0]
// but it would fail if the matrix is empty
int largestCol = 0;
int largestRow = 0;
int?[] largestPerColumn = new int?[numOfCols]; // You could also populate this with
// the values from the first row but
// it would fail if there are no rows
int[] largestColumnRow = new int[numOfCols];
for (int i = 0; i < numOfRows; i++)
{
for (int j = 0; j < numOfCols; i++)
{
if (largestSoFar < myMatrix[i,j])
{
largestSoFar = myMatrix[i,j];
largestCol = j;
largestRow = i;
}
if (largestPerColumn[j] < myMatrix[i,j])
{
largestPerColumn[j] = myMatix[i,j];
largestColumnRow[j] = i;
}
}
}
// largestSoFar is the biggest value in the whole matrix
// largestCol and largestRow is the column and row of the largest value in the matrix
// largestPerColumn[j] is the largest value in the jth column
// largestColumnRow[j] is the row of the largest value of the jth column
If you do need to capture all the "maxima" (for want of a better word, because that's not really what you are doing) in a column, you could just change the above code to something like this:
int? largestSoFar = null; // you could populate this with myMatrix[0,0]
// but it would fail if the matrix is empty
int largestCol = 0;
int largestRow = 0;
int?[] largestPerColumn = new int?[numOfCols]; // You could also populate this with
// the values from the first row but
// it would fail if there are no rows
List<int>[] largestColumnRow = new List<int>[numOfCols];
for (int i = 0; i < numOfRows; i++)
{
for (int j = 0; j < numOfCols; i++)
{
if (largestSoFar < myMatrix[i,j])
{
largestSoFar = myMatrix[i,j];
largestCol = j;
largestRow = i;
}
if (largestPerColumn[j] < myMatrix[i,j])
{
largestPerColumn[j] = myMatix[i,j];
largestColumnRow[j].Add(i);
}
}
}
// Now largestColumnRow[j] gives you a list of all the places where you found a larger
// value for the jth column

Converting a bitmap to an matrix

I've been trying to convert a bitmap object to a matrix of int.
I drew the letter 'C' in paint on a blank white sheet and the program was supposed to initialize the arr in place (x,y) with '0' if the pixel in the Bitmap object in the same position (x,y) is white and correspondingly '1' if it were a black pixel.
I wrote the following code:
static void Main(string[] args)
{
Bitmap arie = new Bitmap(#"C:\Users\User\Desktop\letter.bmp");
object [,] arr = new object[arie.Width, arie.Height];
int min=1000,counter=1;
for (int i = 0; i < arr.GetLength(0) - 1; i++)
{
for (int j = 0; j < arr.GetLength(1) - 1; j++)
{
if (arie.GetPixel(i, j).ToArgb() == Color.White.ToArgb())
{
arr[i, j] = 0;
}
else
arr[i, j] = 1;
}
}
for (int i = 1; i < arr.GetLength(0) - 2; i++)
{
for (int j = 1; j < arr.GetLength(1) - 2; j++)
{
Console.Write(arr[i, j]);
}
Console.WriteLine();
}
}
at the end the letter 'C' that I drew came out like this:
http://teachers.web.cern.ch/teachers/archiv/hst2000/teaching/expt/wavesand/image39.gif
Can anyone seem to recognize the issue?

The main issue is that when you are outputting the array to the console, you are traversing it by columns, rather than rows. That is why your array appears to be rotated by 90 degrees.
If in the second looping portion (where you are outputting to the console), you swap the inner and outer loops, you should see better results.
I also agree with the other comments that:
1) If you want better performance, use the LockBits/UnlockBits method of accessing the Bitmap
2) Your loop indexing is out of whack
If you use x and y instead of i and j, it will help you recognize when you're making these kind of mistakes.
Try this:
static void Main(string[] args)
{
Bitmap arie = new Bitmap(#"C:\Users\User\Desktop\letter.bmp");
object [,] arr = new object[arie.Width, arie.Height];
int min=1000,counter=1;
for (int i = 0; i < arie.Width; i++)
{
for (int j = 0; j < arie.Height; j++)
{
if (arie.GetPixel(i, j).ToArgb() == Color.White.ToArgb())
{
arr[i, j] = 0;
}
else
arr[i, j] = 1;
}
}
for (int y = 0; y < arie.Height; y++)
{
for (int x = 0; x < arie.Width; x++)
{
Console.Write(arr[x, y]);
}
Console.WriteLine();
}
}