Binary image (np.int dtype) become all zero after resizing

[Guodonggogo]

Description

I would like resize a binary image (dtype='np.int') using transform.resize, but its output becomes all zeros. But if I change the input image type to 'np.bool_', the output is correct.

Way to reproduce

import numpy as np
import matplotlib.pyplot as plt
from skimage import io, transform

path_image = './mask.png'
label = io.imread(path_image)
label0 = np.array(label>0, np.int)
label1 = np.array(label>0, np.bool_)

fig = plt.figure()
ax1 = fig.add_subplot(131)  
ax2 = fig.add_subplot(132)  
ax3 = fig.add_subplot(133)  
result0 = transform.resize(label0,(288,288), mode='constant', order=0, preserve_range=False)
result1 = transform.resize(label1,(288,288), mode='constant', order=0, preserve_range=False)

ax1.imshow(label*255, vmin=0, vmax=255)
ax2.imshow(result0*255, vmin=0, vmax=255)
ax3.imshow(result1*255, vmin=0, vmax=255)
plt.show()

print('integer', np.max(result0))
print('bool', np.max(result1))

Place the full code we need to recreate your issue here

upload all necessary images to github too!

Version information

# Paste the output of the following python commands
from __future__ import print_function
import sys; print(sys.version)
import platform; print(platform.platform())
import skimage; print(f'scikit-image version: {skimage.__version__}')
import numpy; print(f'numpy version: {numpy.__version__}')

3.7.4 (default, Aug 13 2019, 20:35:49)
[GCC 7.3.0]
Linux-5.4.0-77-generic-x86_64-with-debian-buster-sid
scikit-image version: 0.17.2
numpy version: 1.16.1

# your output here

integer 5.421010862427522e-20
bool 1.0

"preserve_range = True" has same issue
``

`

[mkcor]

Hi @Guodonggogo,

Thanks for reaching out!

It looks like the image didn't upload properly in your post. Would you mind re-uploading it?

Thanks

[mkcor]

Ok, this one I can see, I guess it's the figure with 3 subplots:

[grlee77]

does it work for you with the np.int image if you set anti_aliasing=False when calling resize? (possibly related to #4292 which was fixed by automatically disabling anti-aliasing when the input has bool type. The same automatic disabling will not occur in your case since the type is int rather than bool)

[Guodonggogo]

@grlee77 Thanks a lot for replying. I tried result0 = transform.resize(label0,(288,288), mode='constant', order=0, preserve_range=False, anti_aliasing=False)
But it doesn't help.

[rfezzani]

OK, with current dev version @grlee77 answer does the trick:

import imageio
import numpy as np
import matplotlib.pyplot as plt
from skimage.transform import resize

label = imageio.imread("/tmp/mask.png")

label0 = np.array(label > 0, np.int)
label1 = np.array(label > 0, np.bool_)

result0 = resize(label0, (288, 288), mode='constant', order=0,
                 preserve_range=False, anti_aliasing=False)
result1 = resize(label1, (288, 288), mode='constant', order=0,
                 preserve_range=False, anti_aliasing=False)

fig, (ax0, ax1, ax2) = plt.subplots(1, 3, figsize=(8, 4))

ax0.imshow(label)
ax1.imshow(result0)
ax2.imshow(result1)
fig.tight_layout()
plt.show()

@Guodonggogo, I edited your report message to correctly display your images 😉

[Guodonggogo]

@rfezzani Thank you a lot. So, the result above is from 0.19.0.dev0, is it?

[rfezzani]

@rfezzani Thank you a lot. So, the result above is from 0.19.0.dev0, is it?

Yes, exactly! I hope you don't mind @Guodonggogo if I convert this issue to discussion. Thank you again for your interest.