I have a multi-class segmentation mask eg.
[1 1 1 2 2 2 2 3 3 3 3 3 3 2 2 2 2 4 4 4 4 4 4]
And going to need to get binary segmentation masks for each value
i.e.
[1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
[0 0 0 1 1 1 1 0 0 0 0 0 0 1 1 1 1 0 0 0 0 0 0]
[0 0 0 0 0 0 0 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0]
[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1]
Any elegant numpy way to do this?
Ideally an example, where I can set 0 and 1 to other values, if I have to.
Just do "==" as this
import numpy as np
a = np.array([1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 2, 2, 2, 2, 4, 4, 4, 4, 4, 4])
mask1 = (a==1)*5
mask2 = (a==2)*5
mask3 = (a==3)*5
mask4 = (a==4)*5
for mask in [mask1,mask2,mask3,mask4]:
print(mask)
This gives
[5 5 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
[0 0 0 5 5 5 5 0 0 0 0 0 0 5 5 5 5 0 0 0 0 0 0]
[0 0 0 0 0 0 0 5 5 5 5 5 5 0 0 0 0 0 0 0 0 0 0]
[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 5 5 5 5 5 5]
You can manipulate the masks further in the same manner, i. e.
mask1[mask1==0] = 3
Native python approach:
You can use comprehension and get the equality values for each unique value using set(<sequence>), then convert the boolean to int to get 0,1 values.
>>> ls =[1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 2, 2, 2, 2, 4, 4, 4, 4, 4, 4]
>>> {v:[int(v==i) for i in ls] for v in set(ls)}
{1: [1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
2: [0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0],
3: [0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
4: [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1]}
Numpy approach:
Get the unique values for the list using np.unique then expand the axis and transpose the array, then expand the axis for the list also and repeat it n times where n is the number of unique values, finally do the equality comparison and convert it to integer type:
import numpy as np
ls = [1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 2, 2, 2, 2, 4, 4, 4, 4, 4, 4]
uniques = np.expand_dims(np.unique(ls), 0).T
result = (np.repeat(np.expand_dims(ls, 0), uniques.shape[0], 0)==uniques).astype(int)
OUTPUT:
print(result)
[[1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
[0 0 0 1 1 1 1 0 0 0 0 0 0 1 1 1 1 0 0 0 0 0 0]
[0 0 0 0 0 0 0 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0]
[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1]]
You can build the mask using np.arange and .repeat() and then use broadcasting and the == operator to generate the arrays:
a = np.array([1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 2, 2, 2, 2, 4, 4, 4, 4, 4, 4])
mask = np.arange(a.min(), a.max()+1).repeat(a.size).reshape(-1, a.size)
a_masked = (a == m).astype(int)
print(a_masked.shape) # (4, 23)
print(a_masked)
# [[1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
# [0 0 0 1 1 1 1 0 0 0 0 0 0 1 1 1 1 0 0 0 0 0 0]
# [0 0 0 0 0 0 0 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0]
# [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1]]
Setting 0 and 1 to other values can be done via normal indexing:
a_masked[a_masked == 0] = 7
a_masked[a_masked == 1] = 42
