how to get equivalent result

[lfz]

Thanks for the fast fix for the padding issue!

I found the definition of "geodesic distance transform (GDT)" seems different from what I understand.

I need a function like "bwdistgeodesic" in (matlab), in this senario, it is very close to the traditional distance transform, which is also used for binary image. The only difference is that, the distance calculation cannot transmit through the "0" values in GDT.

but when I try FastGeodis, I found the basic logic differs from "bwdistgeodesic". the distance value only grows when face image gradient

import numpy as np
import torch
import FastGeodis
BW = np.array([[1, 1, 1, 1, 1, 1, 1, 1, 1, 11],
                            [1, 1, 1, 1, 1, 1, 0, 0, 1, 1],
                            [1, 1, 1, 1, 1, 1, 0, 0, 1, 1],
                            [1, 1, 1, 1, 1, 1, 0, 0, 1, 1],
                            [0, 0, 0, 0, 0, 1, 0, 0, 1, 0],
                            [0, 0, 0, 0, 1, 1, 0, 1, 1, 0],
                            [0, 1, 0, 0, 1, 1, 0, 0, 0, 0],
                            [0, 1, 1, 1, 1, 1, 1, 0, 1, 0],
                            [0, 1, 1, 0, 0, 0, 1, 1, 1, 0],
                            [0, 0, 0, 0, 1, 0, 0, 0, 0, 0]])

C = [1, 2, 3, 3, 3]
R = [3, 3, 3, 1, 2]
seed_img = np.zeros_like(BW)
seed_img[C,R] = 1
v = 1e10
# lamb = 0.0 (Euclidean) or 1.0 (Geodesic) (or (0.0, 1.0) (mixture)
lamb = 1.0
iterations = 2
geodesic_dist = FastGeodis.generalised_geodesic2d(
                        torch.from_numpy(BW).cuda()[None,None ].float(),
                        1-torch.from_numpy(seed_img).cuda()[None,None],
                        v, lamb, iterations)

but i get the following result:

what I expect is:

as far as I understand, this lib is designed to use image gradient as energy barrier, is there a simple way for me to reproduce the expected result?

[lfz]

I figured out a walkaround:

1. do traditional 3d edt firstly (edt)
2. set the masked area of edt to a large value, e.g. edt[mask] = 1e4
3. then we use generalised_geodesic, set the result mask to nan or others

but I think this is not a good way to do so, is it possible to use more elegant method and incorporate this function in the lib

besides, I also want to get the nearest seed point for each point in the image, is it possible

[lfz]

In my scenario, it is close to a watershed algorithm, I got several seed points, and what to segment the binary mask to different basins that bound to those seed points. So I want the function can return the nearest seed point index