feature.peak_local_max changes input image

[odoublewen]

Description

Under certain conditions, the input image changes after calling feature.peak_local_max. This behavior is evident in scikit-image 0.18.0 and 0.18.1. It is not present in 0.17.2

Way to reproduce

Input data files attached:
im_int.txt
labels_int.txt

Script:

import skimage
from skimage import feature
import numpy as np

print("skimage version")
print(skimage.__version__)

im = np.genfromtxt('im_int.txt')
labels = np.genfromtxt('labels_int.txt', dtype=int)

print("image before calling feature.peak_local_max")
print(im)
print(np.min(im))

peaks = feature.peak_local_max(im, labels=labels, num_peaks_per_label=1, exclude_border=False)

print("image after calling feature.peak_local_max")
print(im)
print(np.min(im))

Output from both 0.17.2 and 0.18.1:

$ python find_peaks.py 
skimage version
0.17.2
image before calling feature.peak_local_max
[[1145. 1136. 1041. ... 1055.  999. 1153.]
 [1115. 1107. 1073. ... 1001. 1060. 1114.]
 [1091. 1065. 1069. ...  927. 1124. 1109.]
 ...
 [1110. 1168.  908. ... 1149.  917.  953.]
 [1045. 1104.  937. ... 1057. 1094.  916.]
 [ 793.  954. 1128. ... 1013. 1018.  983.]]
793.0
image after calling feature.peak_local_max
[[1145. 1136. 1041. ... 1055.  999. 1153.]
 [1115. 1107. 1073. ... 1001. 1060. 1114.]
 [1091. 1065. 1069. ...  927. 1124. 1109.]
 ...
 [1110. 1168.  908. ... 1149.  917.  953.]
 [1045. 1104.  937. ... 1057. 1094.  916.]
 [ 793.  954. 1128. ... 1013. 1018.  983.]]
793.0


$ python find_peaks.py 
skimage version
0.18.1
image before calling feature.peak_local_max
[[1145. 1136. 1041. ... 1055.  999. 1153.]
 [1115. 1107. 1073. ... 1001. 1060. 1114.]
 [1091. 1065. 1069. ...  927. 1124. 1109.]
 ...
 [1110. 1168.  908. ... 1149.  917.  953.]
 [1045. 1104.  937. ... 1057. 1094.  916.]
 [ 793.  954. 1128. ... 1013. 1018.  983.]]
793.0
image after calling feature.peak_local_max
[[ 1.14500000e+003  1.13600000e+003 -1.79769313e+308 ...  1.05500000e+003
   9.99000000e+002  1.15300000e+003]
 [ 1.11500000e+003  1.10700000e+003 -1.79769313e+308 ...  1.00100000e+003
   1.06000000e+003  1.11400000e+003]
 [ 1.09100000e+003  1.06500000e+003  1.06900000e+003 ...  9.27000000e+002
   1.12400000e+003  1.10900000e+003]
 ...
 [ 1.11000000e+003  1.16800000e+003 -1.79769313e+308 ...  1.14900000e+003
   9.17000000e+002  9.53000000e+002]
 [ 1.04500000e+003 -1.79769313e+308 -1.79769313e+308 ...  1.05700000e+003
   1.09400000e+003  9.16000000e+002]
 [ 7.93000000e+002 -1.79769313e+308 -1.79769313e+308 ...  1.01300000e+003
   1.01800000e+003  9.83000000e+002]]
-1.7976931348623157e+308

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("scikit-image version: {}".format(skimage.__version__))
import numpy; print("numpy version: {}".format(numpy.__version__))

# your output here

3.8.6 (default, Nov  5 2020, 15:37:44) 
[Clang 11.0.0 (clang-1100.0.33.16)]
macOS-10.15.7-x86_64-i386-64bit
scikit-image version: 0.18.1
numpy version: 1.20.2

FYI @rfezzani - Somewhat similar to #5047

[rfezzani]

Thank you @odoublewen for reporting, I will investigate ASAP 😉

[memoryleine]

This seems similar to #5235 as well? I remember because I ran into the same problem recently and discovered that issue when I was trying to figure it out.

[rfezzani]

@memoryleine, I can't reproduce the bug with current skimage dev version:

>>> import skimage
>>> from skimage import feature
>>> import numpy as np
>>> 
>>> print("skimage version")
skimage version
>>> print(skimage.__version__)
0.19.0.dev0
>>> 
>>> im = np.genfromtxt('im_int.txt')
>>> labels = np.genfromtxt('labels_int.txt', dtype=int)
>>> 
>>> print("image before calling feature.peak_local_max")
image before calling feature.peak_local_max
>>> print(im)
[[1145. 1136. 1041. ... 1055.  999. 1153.]
 [1115. 1107. 1073. ... 1001. 1060. 1114.]
 [1091. 1065. 1069. ...  927. 1124. 1109.]
 ...
 [1110. 1168.  908. ... 1149.  917.  953.]
 [1045. 1104.  937. ... 1057. 1094.  916.]
 [ 793.  954. 1128. ... 1013. 1018.  983.]]
>>> print(np.min(im))
793.0
>>> 
>>> peaks = feature.peak_local_max(im, labels=labels, num_peaks_per_label=1, exclude_border=False)
>>> 
>>> print("image after calling feature.peak_local_max")
image after calling feature.peak_local_max
>>> print(im)
[[1145. 1136. 1041. ... 1055.  999. 1153.]
 [1115. 1107. 1073. ... 1001. 1060. 1114.]
 [1091. 1065. 1069. ...  927. 1124. 1109.]
 ...
 [1110. 1168.  908. ... 1149.  917.  953.]
 [1045. 1104.  937. ... 1057. 1094.  916.]
 [ 793.  954. 1128. ... 1013. 1018.  983.]]
>>> print(np.min(im))
793.0

I will investigate more concerning v18 😉 .

[rfezzani]

OK, as @memoryleine pointed out 👏, this bug is already reported in #5235 and it is fixed by #5236 😉. I will close this issue as it is already fixed. Thank you again @odoublewen for reporting.