MultiMasks Augmentation - Issue with torch.from_numpy() - Abumentation

[FrsECM]

Describe the bug

Hello !
I have a problem when i use HorizontalFlip combined with multiple masks and ToTensorV2 augmentation.

To Reproduce

You can use this piece of code to reproduce :

import albumentations as A
from albumentations.pytorch import ToTensorV2
import cv2
import numpy as np

transform = A.Compose([
        A.HorizontalFlip(p=1),
        A.ToFloat(max_value=255),
        ToTensorV2()
    ],is_check_shapes=False)


image = 255*np.ones((3096,2048,3),dtype=np.uint8)
masks = [np.ones((3096,2048),dtype=np.uint8) for i in range(2)]

transformed = transform(image=image,masks=masks)

Steps to reproduce the behavior:

1. Python 3.10 - Albumentation 1.14.13
2. Sample code that produces the bug
3. Any error messages or incorrect outputs.

Expected behavior

I expect the augmentation to be applied without error

Actual behavior

I have an exception :

Traceback (most recent call last):
  File "/home/default/miniconda/envs/domf_iris2/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/home/default/miniconda/envs/domf_iris2/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/home/f296849/.vscode-server/extensions/ms-python.debugpy-2024.6.0/bundled/libs/debugpy/adapter/../../debugpy/launcher/../../debugpy/__main__.py", line 39, in <module>
    cli.main()
  File "/home/f296849/.vscode-server/extensions/ms-python.debugpy-2024.6.0/bundled/libs/debugpy/adapter/../../debugpy/launcher/../../debugpy/../debugpy/server/cli.py", line 430, in main
    run()
  File "/home/f296849/.vscode-server/extensions/ms-python.debugpy-2024.6.0/bundled/libs/debugpy/adapter/../../debugpy/launcher/../../debugpy/../debugpy/server/cli.py", line 284, in run_file
    runpy.run_path(target, run_name="__main__")
  File "/home/f296849/.vscode-server/extensions/ms-python.debugpy-2024.6.0/bundled/libs/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_runpy.py", line 321, in run_path
    return _run_module_code(code, init_globals, run_name,
  File "/home/f296849/.vscode-server/extensions/ms-python.debugpy-2024.6.0/bundled/libs/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_runpy.py", line 135, in _run_module_code
    _run_code(code, mod_globals, init_globals,
  File "/home/f296849/.vscode-server/extensions/ms-python.debugpy-2024.6.0/bundled/libs/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_runpy.py", line 124, in _run_code
    exec(code, run_globals)
  File "/home/f296849/code/2024/domf_iris2/notebooks/sandbox/debug_albumentation_1.4.13.py", line 27, in <module>
    transformed = transform(image=image,masks=masks)
  File "/home/default/miniconda/envs/domf_iris2/lib/python3.10/site-packages/albumentations/core/composition.py", line 307, in __call__
    data = t(**data)
  File "/home/default/miniconda/envs/domf_iris2/lib/python3.10/site-packages/albumentations/core/transforms_interface.py", line 123, in __call__
    return self.apply_with_params(params, **kwargs)
  File "/home/default/miniconda/envs/domf_iris2/lib/python3.10/site-packages/albumentations/core/transforms_interface.py", line 141, in apply_with_params
    res[key] = target_function(arg, **params)
  File "/home/default/miniconda/envs/domf_iris2/lib/python3.10/site-packages/albumentations/pytorch/transforms.py", line 52, in apply_to_masks
    return [self.apply_to_mask(mask, **params) for mask in masks]
  File "/home/default/miniconda/envs/domf_iris2/lib/python3.10/site-packages/albumentations/pytorch/transforms.py", line 52, in <listcomp>
    return [self.apply_to_mask(mask, **params) for mask in masks]
  File "/home/default/miniconda/envs/domf_iris2/lib/python3.10/site-packages/albumentations/pytorch/transforms.py", line 49, in apply_to_mask
    return torch.from_numpy(mask)
ValueError: At least one stride in the given numpy array is negative, and tensors with negative strides are not currently supported. (You can probably work around this by making a copy of your array  with array.copy().) 

Screenshots

You can see bellow that it's working with v1.4.12

With v1.4.13, it is no longer working :

Additional context

I suspect this commit to be the cause of the problem :

• bcaf7d6

Workarround :

import albumentations as A
from albumentations.pytorch import ToTensorV2
import numpy as np
import torch
from albumentations.core.types import NUM_MULTI_CHANNEL_DIMENSIONS

class ToTensor(ToTensorV2):
    def apply_to_mask(self, mask: np.ndarray, **params: A.Any) -> torch.Tensor:
        if self.transpose_mask and mask.ndim == NUM_MULTI_CHANNEL_DIMENSIONS:
            mask = mask.transpose(2, 0, 1)
        return torch.from_numpy(np.ascontiguousarray(mask))

[ternaus]

Thanks, I can reproduce the bug.

Will take a look ASAP.

[ternaus]

In [4]: import albumentations as A
   ...: from albumentations.pytorch import ToTensorV2
   ...: import cv2
   ...: import numpy as np
   ...:
   ...: transform = A.Compose([
   ...:         A.HorizontalFlip(p=1),
   ...:         A.ToFloat(max_value=255),
   ...:         ToTensorV2()
   ...:     ],is_check_shapes=False)
   ...:
   ...:
   ...: image = 255*np.ones((3096,2048,3),dtype=np.uint8)
   ...: masks = [np.ones((3096,2048),dtype=np.uint8) for i in range(2)]
   ...:
   ...: transformed = transform(image=image,masks=masks)

In [5]: A.__version__
Out[5]: '1.4.15'

Could you please check with the latest version?

[FrsECM]

For me it's not fixed with the last version :

import albumentations as A
from albumentations.pytorch import ToTensorV2
import cv2
import numpy as np
import torch
from albumentations.core.types import NUM_MULTI_CHANNEL_DIMENSIONS

################### WORKARROUND THAT WORK
class ToTensor(ToTensorV2):
    def apply_to_mask(self, mask: np.ndarray, **params: A.Any) -> torch.Tensor:
        if self.transpose_mask and mask.ndim == NUM_MULTI_CHANNEL_DIMENSIONS:
            mask = mask.transpose(2, 0, 1)
        return torch.from_numpy(np.ascontiguousarray(mask))

transform_buggy = A.Compose([
        A.HorizontalFlip(p=1),
        A.ToFloat(max_value=255),
        ToTensorV2()
    ],is_check_shapes=False)


transform_fixed = A.Compose([
        A.HorizontalFlip(p=1),
        A.ToFloat(max_value=255),
        ToTensor()
    ],is_check_shapes=False)

image = 255*np.ones((3096,2048,3),dtype=np.uint8)
masks = [np.ones((3096,2048),dtype=np.uint8) for i in range(2)]

print(A.__version__)
try:
    transformed = transform_fixed(image=image,masks=masks)
except:
    print('transform_fixed failed')
try:
    transformed_buggy = transform_buggy(image=image,masks=masks)
except:
    print('transform_buggy failed')
print('The end')