Encountering problem when augmented with mask.

🐛 Bug

The augmented mask is not match to the augmented image.

To Reproduce

Steps to reproduce the behavior:

aug = A.Compose([A.OneOf([A.HorizontalFlip(p=0.5), A.VerticalFlip(p=0.5), A.Transpose(p=0.5)], p=1),
      A.RandomRotate90(p=0.5),
      A.GridDistortion(p=0.5)])

augmented  = aug(image=image, mask=label)
image = augmented['image']
label= augmented['mask']

Expected behavior

Environment

• Albumentations version (e.g., 0.1.8): 0.4.6
• Python version (e.g., 3.7): python 3.7.4
• OS (e.g., Linux): Ubuntu 18.04.3 LTS
• How you installed albumentations (conda, pip, source): conda
• Any other relevant information:

Additional context

• original images with mask

• wrong augmented result, the mask is not match to the image

[super233]

Sorry, the problem was caused by I called __getitem__() twice to display the result, it’s a stupid problem, thanks for your reply. 😞

[Dipet]

I again can not reproduce the problem

from torch.utils.data import Dataset
from torchvision import transforms
import numpy as np
import albumentations as A
import matplotlib.pyplot as plt
import random


class CrossMoDADataSet(Dataset):
    def __init__(self):
        self.augment = self.build_augment()
        self.to_tensor_and_normalize = transforms.Compose([transforms.ToTensor()])

    def __len__(self):
        return 1

    def __getitem__(self, index):
        image = np.load("D:\\crossmoda_22_ceT1_40.npy")
        label = np.load("D:\\crossmoda_22_Label_40.npy")

        image = (image / np.max(image) * 255).astype(np.uint8)  # scale the value to [0, 255]
        label = label.astype(np.uint8)
        image_orig = image.copy()
        label_orig = label.copy()

        if self.augment is not None:
            augmented = self.augment(image=image, mask=label)
            image = augmented['image']
            label = augmented['mask']

        return image, label, image_orig, label_orig

    def build_augment(self):
        return A.Compose([A.OneOf([A.HorizontalFlip(p=0.5),
                                   A.VerticalFlip(p=0.5),
                                   A.Transpose(p=0.5)], p=1),
                          A.RandomRotate90(p=0.5),
                          A.GridDistortion(p=0.5)])


seed = 0
random.seed(seed)
np.random.seed(seed)
a, b, c, d = CrossMoDADataSet()[0]

plt.subplot(221, title="original")
plt.imshow(c, vmin=0, vmax=255, cmap="gray")
plt.subplot(222, title="original")
plt.imshow(d, vmin=0, vmax=1, cmap="gray")

plt.subplot(223, title="result")
plt.imshow(a, vmin=0, vmax=255, cmap="gray")
plt.subplot(224, title="result")
plt.imshow(b, vmin=0, vmax=1, cmap="gray")

plt.show()