🐛 Bug

I was experimenting with albumentations functions when I found this bug. If "ignore_values" parameter in "CropNonEmptyMaskIfExists" isn't explicitly assigned a value, the transformation may modify the masks passed to it.

To Reproduce

Steps to reproduce the behavior:

import albumentations as A
import numpy as np
a = numpy.array([[1,2],[3,4]])
b = numpy.array([[5,6],[7,8]])
img = numpy.array([[11,12],[13,14]])
transform = A.augmentations.crops.transforms.CropNonEmptyMaskIfExists(1,2)
out = transform(image = img, masks = [a,b])

Expected behavior

According to the documentation, I assumed the transformation will work as if ignore_values=[0] and simply give a randomly cropped image and masks. However, the values of variable "out" are:

{'image': array([[13, 14]]), 'masks': [array([[ 7, 12]]), array([[7, 8]])]}

Meanwhile, the values of "a" are now:

array([[ 5,  6],
       [ 7, 12]])

Setting "ignore_channels" parameter wouldn't change this behavior. However, if "mask" was passed instead of "masks", the transformation would work as it's supposed to.

Environment

• Albumentations version (e.g., 0.1.8): 1.1.0
• Python version (e.g., 3.7): python 3.9.6
• OS (e.g., Linux): windows 10
• How you installed albumentations (conda, pip, source): pip