feature.peak_local_max modifies input image

[simone-codeluppi]

Description

Hi
I am processing a float image (linked below) to detect peaks.
When i call feature.peak_local_max without defining labels everything is fine. If I define labels the call to feature.peak_local_max causes a change in the intensity values (adding/changing the value of some pixels to 1.7976931348623157e+308) of the image processed. I can replicate the issue with skimage 0.18.0 and 0.18.1 however everything works fine with skimage 0.17.2. Please let me know if I missed anything because I cannot wrap my head around the issue. Thanks!

File to reproduce the issue

img_path_dropbox = get image from here

Way to reproduce

from skimage import feature, measure
from scipy import ndimage as nd

def test(img):
    print(f'min value original: {img.min()}')
    thr = 2.208
    img_mask = img>thr
        
    labels = nd.label(img_mask)[0]
            
    properties = measure.regionprops(labels)
                
    for ob in properties:
        if ob.area< 10 or ob.area>600:
            img_mask[ob.coords[:,0],ob.coords[:,1]]=0

    labels = nd.label(img_mask)[0]
    properties = measure.regionprops(labels,intensity_image=img)

    selected_peaks = feature.peak_local_max(img, min_distance=5, 
                            threshold_abs=thr, exclude_border=True, 
                            footprint=None,num_peaks_per_label=1)

    print(f'min value after peak calling no labels: {img.min()}')

    selected_peaks = feature.peak_local_max(img, min_distance=5, 
                            threshold_abs=thr, exclude_border=True, 
                            footprint=None, labels=labels,num_peaks_per_label=1)
    
    print(f'min value after peak calling using labels: {img.min()}')
    return selected_peaks


img, meta = pickle.load(open(path-to-img ,'rb'))
pk = test(img, thr)

# RESULT 
# min value original: 0.2976309153266585
# min value after peak calling no labels: 0.2976309153266585
# min value after peak calling using labels: -1.7976931348623157e+308

Version information

3.8.2 (default, May  7 2020, 20:00:49)
[GCC 7.3.0]
Linux-3.10.0-957.1.3.el7.x86_64-x86_64-with-glibc2.10
scikit-image version: 0.18.0
# same error in scikit-image version: 0.18.1
# everything works fine in version 0.17.2
numpy version: 1.19.4

[rfezzani]

Thank you @simone-codeluppi for the report. I confirm the bug, I can reproduce 😖