use_pin_hole (x,y) vs (y,x)

I had problems with the use_pin_hole option. It worked fine on the test data but not on some real world data (kidney segmentations)
I figured the test data was symmetric on y=x.

The problem seems to come from image_help.py, def draw_line_upwards_from_point.

I had to make some changes to (x,y) to (y,x) and back, to get it working on my data.

def draw_line_upwards_from_point(
mask: np.ndarray, start, fill_value: int
) -> np.ndarray:
line_width = 2
end = (start[1], start[0]-1) # np.ndarray and opencv expects (x,y) vs (y,x)
#start = (end[0], end[1])
mask = mask.astype(np.uint8) # Type that OpenCV expects
# Draw one point at a time until we hit a point that already has the desired value
while mask[end] != fill_value:
#cv.line(mask, start, end, fill_value, line_width) # skipped, no need to draw each step
#print("debug",start,end,mask[end],fill_value)
# Update end to the next position
#start = end
end = (end[0]- line_width, end[1])
cv.line(mask, start, (end[1],end[0]), fill_value, line_width) # np.ndarray and opencv expects (x,y) vs (y,x)
#plt.imshow(mask)
#plt.show()
return mask.astype(bool)

The issue points to image_help.py and draw_line_upwards_from_point; first inspect the coordinate conventions used by the mask and OpenCV calls, then reproduce the use_pin_hole case with nonsymmetric data. Done means the function handles (x,y) and (y,x) consistently and use_pin_hole works for kidney segmentations without manual coordinate swaps.