colored image exported as greyscale

[mhoibo]

Hi, I am using fast to extract patches from several images (.svs-format) , using the fast patchGenerator() and fast.ImageExporter(). I have had issues with a few images (3/21) that when I extract patches from them, the patches are grey (not colored). I am not sure if this is really a fast issue, but I would be grateful for any pointers.

I used this fast wheel (https://github.com/smistad/FAST/actions/runs/12353873040, numpy 1.26.4).

The same color issue happened for the "problematic images" when trying to display with matplotlib:

path = 'path/to/image/'
importer = fast.WholeSlideImageImporter.create(path)
image_pyramid = importer.runAndGetOutputData()
im_ = image_pyramid.getAccess(fast.ACCESS_READ).getLevelAsImage(level=1)
plt.imshow(im_)
plt.show()

print("depth: ", im_.getDepth()) -> 1
print("dimensions, ", im_.getDimensions()) -> 2
print("Nbr channels: ", im_.getNrOfChannels()) -> 3

When opening the original full images with qupath they look good, but the extracted patches are grey when opened in qupath (though pixel type still says uint8 (rgb)). The patches from the remaining 18 images are good when opened in qupath.

I also tried to install the latest stable fast version, which also did not work. However, when trying on a windows-machine (locally) it did work.

Does anyone have any ideas to what the issue might be?

[smistad]

Opening the same image gives gray image on linux, but works fine on windows?

[mhoibo]

yes, correct, when trying the code above on linux and windows, it worked on windows but not linux. And it only applied to a few images.

On windows, this wheel was used: https://github.com/smistad/FAST/actions/runs/12085939701 with numpy 1.24.3

[smistad]

At the moment a different openslide version is used on windows and linux, so that might be the issue

[mhoibo]

Okay, thanks for the quick reply :)
(On linux the rgb-channels all seems to be equal, while they are not on windows)

[mhoibo]

Hi! I am not sure its relevant, or helps to understand the problem, but I tested to load one of the greyscale images with pyvips on linux and then i can extract a patch which is correctly colored:

import pyvips
import numpy as np
import matplotlib.pyplot as plt

path = '/path/to/images/image.svs'
image = pyvips.Image.new_from_file(path, access='random')
patch = image.crop(40000, 10000, 5000, 5000)
patch_np = np.asarray(patch)
plt.imshow(patch_np)
plt.show()