如何堆叠多个二进制掩码以创建用于多类分割的单个掩码？

Question

So I have 5 classes for segmentation (excluding background) and all the masks are binary masks in different folders. How do I combine the binary masks to create a single mask for all classes for multiclass segmentation?

I tried by using 2 classes in different color channels, but since there are only 3 color channels I could only incorporate the background and 2 classes in a single mask.

python code:

new_img = np.zeros((height, width,3))

new_img[:,:,0] = new_img[:,:,0] + class1[:,:,0]

new_img[:,:,1] = new_img[:,:,1] + class2[:,:,0]

Is there any other way to do this? and could I use the binary masks for multi-class segmentation without combining them?

Thanks

Answer 1

You can combine the masks by making a mask with 6 channels:

mask = np.stack([mask0, mask1, mask2, mask3, mask4, mask5], axis=-1)

Assuming each maski for i in 0..5 is a 2d array.

If you want to visualize the image, you have to map each class to a color. For example:

@np.vectorize(signature='(6)->(3)')
def mapcolors(x):
    i = np.argmax(x)
    if i == 0:
        return [0.,0.,0.]
    elif i == 1:
        return [0.,0.,1.]
    elif i == 2:
        return [0.,1.,0.]
    elif i == 3:
        return [1.,0.,0.]
    elif i == 4:
        return [0.,1.,1.]
    else:
        return [1.,1.,0.]

image = mapcolors(mask)