Math domain error

[dmerinodel]

Hi,

I'm using this code to process some images from the FLIR i7 camera. I know that theoretically the library doesn't support that camera, but after some handwork with exiftool and images from the camera, I realized that I could use it for extract the thermal_np_array which is enough for me.

Even though I made it work properly for a lot of images, I'm getting a 'math domain error' in the raw2temp module in some cases. Do you know why is this happening? All metadata that raw2temp needs are available so I guess that this shouldn't be happening.

Here you have some examples of the 'troublemakers' :)
Thanks in advance. @Nervengift

[dmerinodel]

Hi,

I kept studying that error and after some 'debugging' in the raw2temp module, I discovered that my thermal_np_array had some 0 values. I'm assuming that these are some hardware bad detections and I'm replacing those 0s with some non zero value from the same row (e.g, the first non zero value). I'm including this piece of code

for row in thermal_np:
    if 0 in row:
        non_zeros = row[row != 0]
        row[row == 0] = non_zeros[0]

Update 18-06-21. That method led to bad coloured images. I ended up adding the following module.

def zeros_filter(matrix, row, col):
    """
    Input: bidimensional np.array, row and column where the 0 lies.
    Output: average of the surrounding values. 
    """
    shape = np.shape(matrix)
    rows = shape[0]
    cols = shape[1]
    
    # Surroundings depends on the position of the element (corners, edges or inner 
    # points). Each 'if/elif' is for each possible position in the matrix. Surrounding
    # non-zero elements are selected for the interpolation and if there isn't any, a
    # non-zero value from the row is taken.
    
    if (row not in {0,rows-1}) and (col not in {0,cols-1}): # inner points
        surr = np.array([matrix[row-1][col-1],matrix[row-1][col],matrix[row][col+1],
                         matrix[row][col-1],matrix[row][col+1],
                         matrix[row+1][col-1],matrix[row+1][col],matrix[row+1][col+1]])
        surr = surr[surr != 0]
        if len(surr) == 0:
            new_value = matrix[row][matrix[row] != 0][0]
        else:
            mean = np.mean(surr)
            new_value = mean

    elif (row == 0): # top row
        if (col == 0):
            surr = np.array([matrix[row][col+1],matrix[row+1][col],
                             matrix[row+1][col+1]])
            surr = surr[surr != 0]
            if len(surr) == 0:
                new_value = matrix[row][matrix[row] != 0][0]
            else:
                mean = np.mean(surr)
                new_value = mean

        elif (col == cols-1):
            surr = np.array([matrix[row][col-1],matrix[row+1][col],
                             matrix[row+1][col-1]])
            surr = surr[surr != 0]
            if len(surr) == 0:
                new_value = matrix[row][matrix[row] != 0][0]
            else:
                mean = np.mean(surr)
                new_value = mean

        else:
            surr = np.array([matrix[row][col-1],matrix[row][col+1],
                             matrix[row+1][col-1],matrix[row+1][col],matrix[row+1][col+1]])
            surr = surr[surr != 0]
            if len(surr) == 0:
                new_value = matrix[row][matrix[row] != 0][0]
            else:
                mean = np.mean(surr)
                new_value = mean
    elif (row == rows-1): # bottom row
        if (col == 0):
            surr = np.array([matrix[row][col+1],matrix[row-1][col],
                             matrix[row][col+1]])
            surr = surr[surr != 0]
            if len(surr) == 0:
                new_value = matrix[row][matrix[row] != 0][0]
            else:
                mean = np.mean(surr)
                new_value = mean
        elif (col == cols-1):
            surr = np.array([matrix[row][col-1],matrix[row-1][col],
                             matrix[row-1][col-1]])
            surr = surr[surr != 0]
            if len(surr) == 0:
                new_value = matrix[row][matrix[row] != 0][0]
            else:
                mean = np.mean(surr)
                new_value = mean
        else:
            surr = np.array([matrix[row-1][col-1],matrix[row-1][col],matrix[row][col+1],
                             matrix[row][col-1],matrix[row][col+1]])
            surr = surr[surr != 0]
            if len(surr) == 0:
                new_value = matrix[row][matrix[row] != 0][0]
            else:
                mean = np.mean(surr)
                new_value = mean
    elif (col == 0): # first column
        surr = np.array([matrix[row-1][col],matrix[row][col+1],
                         matrix[row][col+1],
                         matrix[row+1][col],matrix[row+1][col+1]])
        surr = surr[surr != 0]
        if len(surr) == 0:
            new_value = matrix[row][matrix[row] != 0][0]
        else:
            mean = np.mean(surr)
            new_value = mean
    elif (col == cols-1):  # last column
        surr = np.array([matrix[row-1][col],matrix[row][col-1],
                         matrix[row][col-1],
                         matrix[row+1][col],matrix[row+1][col-1]])
        surr = surr[surr != 0]
        
        if len(surr) == 0:
            new_value = matrix[row][matrix[row] != 0][0]
        else:
            mean = np.mean(surr)
            new_value = mean

    return new_value

I think variable names and comments are clear enough. It is some kind of "convolution". If someone knows how to do it in a easier way I'd appreciate it :)