About Change = np.divide((Changemap + cdMap),2).astype(int)

[haoranwu22]

for sigma in range(101, 202, 50):
    adaptiveThreshold = 2 * filters.gaussian(detectedChangeMapNormalized, sigma)
    cdMapTemp = (detectedChangeMapNormalized > adaptiveThreshold)
    cdMapTemp = morphology.remove_small_objects(cdMapTemp, min_size=objectMinSize)
    cdMap = cdMap | cdMapTemp
cdMap = morphology.binary_closing(cdMap, morphology.disk(3))
cdMap = cdMap.squeeze()

# TODO:是否更改取整方式
Change = np.divide((Changemap + cdMap),2).astype(int)
# Change = np.ceil(np.divide((Changemap + cdMap),2))
Uchange = np.where(Changemap + cdMap==0, 1, 0)
Toclassified = Ones - Change - Uchange

hello, is it reasonable to round down in the source code?
Change = np.divide((Changemap + cdMap),2).astype(int)
or
or Change = np.ceil(np.divide((Changemap + cdMap),2))
Because when experimenting on my own dataset, Change will have a value less than 1.0 after multiple rounds of training, and asype(int) will cause all elements to be rounded down to zero.