Describe the bug

When using the Albumentations package on Google Colab, transformations sometimes work as expected, but other times they fail without any changes to the code. This inconsistency occurs when running the same code multiple times.

To Reproduce

Steps to reproduce the behavior:

Run the same transformation code multiple times on Google Colab.
Observe that some executions succeed while others fail without any apparent reason.

Sample code that produces the bug:

import albumentations as A
import cv2
import matplotlib.pyplot as plt

# Declare an augmentation pipeline
transform = A.Compose([
    A.ToGray(num_output_channels=3, method="weighted_average")
])

# Read an image with OpenCV and convert it to the RGB colorspace
image = cv2.imread("/path/to/image.png")
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)

# Augment an image
transformed = transform(image=image)
transformed_image = transformed["image"]

plt.figure(figsize=(8, 4))
plt.subplot(1, 2, 1)
plt.imshow(image)
plt.title("Original")
plt.axis("off")

plt.subplot(1, 2, 2)
plt.imshow(transformed_image)
plt.title("ToGray")
plt.axis("off")

plt.show()

Expected behavior

The transformations should work consistently across multiple runs without intermittent failures.

Actual behavior

The same transformation code sometimes works and sometimes does not.

Screenshots / GIF

Additional context

When I use this package to generate data for a dataset and run a loop to apply transformations, some images get augmented while others do not.