I am trying to export my NumPy matrix to a .tiff file then to read its contents to verify if it worked. Given the following snippet
import numpy as np
import cv2
data = np.array([[-0.1 for _ in range(128)] for _ in range(128)], dtype=np.float32)
cv2.imwrite('temp.tiff', data)
print('tiff file written')
image = cv2.imread('temp.tiff')
print(image)
returns
tiff file written
[ WARN:0@19.816] global /io/opencv/modules/imgcodecs/src/grfmt_tiff.cpp (629) readData OpenCV TIFF: TIFFRGBAImageOK: Sorry, can not handle images with 32-bit samples
So then I lower down the dtype parameter to dtype=np.float16
data = np.array([[-0.1 for _ in range(128)] for _ in range(128)], dtype=np.float16)
but then I get another error...
, line 50, in export_images_to_tiff
cv2.imwrite('temp.tiff', a)
cv2.error: OpenCV(4.6.0) :-1: error: (-5:Bad argument) in function 'imwrite'
> Overload resolution failed:
> - img data type = 23 is not supported
> - Expected Ptr<cv::UMat> for argument 'img'
[1] 24228 abort python3 main.py
I am honestly going nuts with this feature I'm trying to implement. Is there any alternatives or fix to the above? I read the documentation but that didn't help, and looking online these error codes yield nothing relevant. Do I just assume that the tiff file was successfully written in the first 32-bit pass and call it a day?
cv2.imread() has a second parameter, flags, which specifies how to read the image file. According the the docs, the default value of the flags argument to imread() is this:
IMREAD_COLOR
If set, always convert image to the 3 channel BGR color image.
Source: imread docs, IMREAD_COLOR docs.
By default, OpenCV attempts to coerce a file into 3-channel color. Since you have only a single channel, that's not going to work. However, there is a flag which will work:
image = cv2.imread('temp.tiff', cv2.IMREAD_UNCHANGED)
(Thanks to this answer for pointing me in the right direction.)