MultiplicativeNoise(elementwise=True) fails when image shape is (H,W,1)

[mjkvaak]

Hi,
first of all thanks for one of the most useful ML packages I have found so far! I noticed that albumentations.MultiplicativeNoise fails under these conditions:

• Image is unsqueezed grayscale with shape (H,W,1)
• elementwise parameter has been set to True

Below is the minimal reproducing example:

h,w = 28,28
img = np.zeros((h,w))
tfm_img = albumentations.MultiplicativeNoise(elementwise=True, p=1)(image=img)
print(tfm_img['image'].shape) #prints (28,28) as it should

img = np.zeros((h,w,1))
tfm_img = albumentations.MultiplicativeNoise(elementwise=True, p=1)(image=img)
print(tfm_img['image'].shape) #prints (28,28,28) which is a bug

img = np.zeros((h,w,3))
tfm_img = albumentations.MultiplicativeNoise(elementwise=True, p=1)(image=img)
print(tfm_img['image'].shape) #prints (28,28,3) as it should

Also, if h and w are unequal, the script will crash with ValueError:

h,w = 256,28
img = np.zeros((h,w,1))
tfm_img = albumentations.MultiplicativeNoise(elementwise=True, p=1)(image=img)

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-152-5abd2e633fe7> in <module>
      5 
      6 img = np.zeros((h,w,1))
----> 7 tfm_img = albumentations.MultiplicativeNoise(elementwise=True, p=1)(image=img)

...

ValueError: operands could not be broadcast together with shapes (256,30,1) (256,30) 

[BloodAxe]

Thanks for your report. We will investigate this.