CuPy 14.0 compatibility

python/cucim/src/cucim/skimage/_shared/compat.py

@@ -1,5 +1,6 @@
 """Compatibility helpers for dependencies."""
 
+import cupy as cp
 import numpy as np
 from packaging.version import parse
 
@@ -21,10 +22,9 @@
 
 
 # check CuPy instead of SciPy
-# as of CuPy 13.0, tol is still being used instead of rtol as in latest SciPy
-# CUPY_LT_14 = parse(cp.__version__) < parse("14.0")
+CUPY_LT_14 = parse(cp.__version__) < parse("14.0.0a1")
 
 # Starting in SciPy v1.12, 'scipy.sparse.linalg.cg' keyword argument `tol` is
-# deprecated in favor of `rtol`.
-# As of CuPy 13.0, it is still always using 'tol''
-SCIPY_CG_TOL_PARAM_NAME = "tol"  # if CUPY_LT_14 else "rtol"
+# deprecated in favor of `rtol`. The corresponding change in cupyx.scipy.sparse
+# was made in v14.0
+SCIPY_CG_TOL_PARAM_NAME = "tol" if CUPY_LT_14 else "rtol"

python/cucim/src/cucim/skimage/feature/tests/test_corner.py

@@ -264,8 +264,8 @@ def test_structure_tensor_eigenvalues(dtype):
 
 
 def test_structure_tensor_eigenvalues_3d():
-    cube9 = cp.ones((9,) * 3, dtype=cp.uint8)
-    cube7 = cp.ones((7,) * 3, dtype=cp.uint8)
+    cube9 = cp.ones((9,) * 3, dtype=cp.int16)
+    cube7 = cp.ones((7,) * 3, dtype=cp.int16)
     image = pad(cube9, 5, mode="constant") * 1000
     boundary = (
         pad(cube9, 5, mode="constant") - pad(cube7, 6, mode="constant")

python/cucim/src/cucim/skimage/filters/thresholding.py

@@ -694,6 +694,7 @@ def threshold_li(
     elif callable(initial_guess):
         t_next = initial_guess(image)
     elif cp.isscalar(initial_guess):  # convert to new, positive image range
+        image_min = float(image_min)
         t_next = initial_guess - image_min
         image_max = cp.max(image) + image_min
         if not 0 < t_next < cp.max(image):

python/cucim/src/cucim/skimage/registration/_optical_flow.py

@@ -311,7 +311,7 @@ def _ilk(
     # is the solution of the ndim x ndim linear system
     # A[i, j] * X = b[i, j]
     A = cp.zeros(reference_image.shape + (ndim, ndim), dtype=dtype)
-    b = cp.zeros(reference_image.shape + (ndim,), dtype=dtype)
+    b = cp.zeros(reference_image.shape + (ndim, 1), dtype=dtype)
 
     grid = cp.meshgrid(
         *[cp.arange(n, dtype=dtype) for n in reference_image.shape],
@@ -337,15 +337,15 @@ def _ilk(
             A[..., i, j] = A[..., j, i] = filter_func(grad[i] * grad[j])
 
         for i in range(ndim):
-            b[..., i] = filter_func(grad[i] * error_image)
+            b[..., i, 0] = filter_func(grad[i] * error_image)
 
         # Don't consider badly conditioned linear systems
         idx = abs(cp.linalg.det(A)) < 1e-14
         A[idx] = cp.eye(ndim, dtype=dtype)
         b[idx] = 0
 
         # Solve the local linear systems
-        flow = cp.moveaxis(cp.linalg.solve(A, b), ndim, 0)
+        flow = cp.moveaxis(cp.linalg.solve(A, b)[..., 0], ndim, 0)
 
     return flow
 

python/cucim/src/cucim/skimage/transform/integral.py

@@ -148,10 +148,10 @@ def integrate(ii, start, end):
 
         # CuPy Backend: TODO: check efficiency here
         if scalar_output:
-            S += sign * ii[tuple(corner_points[0])] if (not bad[0]) else 0
+            S += sign * int(ii[tuple(corner_points[0])]) if (not bad[0]) else 0
         else:
             for r in range(rows):
                 # add only good rows
                 if not bad[r]:
-                    S[r] += sign * ii[tuple(corner_points[r])]
+                    S[r] += sign * int(ii[tuple(corner_points[r])])
     return S