feat: update rp factory to use own find obj function

Co-authored-by: Copilot <copilot@github.com>

Author: Teranis

.github/workflows/build_cython_extensions.yml

@@ -2,9 +2,7 @@ name: Build Cython extensions
 
 on:
   push:
-    branches:
-      - main
-    paths:
+paths:
       - "cellacdc/**/*.pyx"
       - "setup.py"
 

cellacdc/precompiled/__init__.py

@@ -6,6 +6,12 @@
     _normalize_spacing,
 )
 import traceback as traceback
+
+try:
+    from .precompiled.regionprops_helper import find_all_objects_2D, find_all_objects_3D
+    _CYTHON_FIND_OBJECTS = True
+except Exception:
+    _CYTHON_FIND_OBJECTS = False
 # WARNING: Developers have already used
 #     7 hrs
 # to optimize this.
@@ -55,23 +61,36 @@ def _acdc_regionprops_factory(
             )
 
     regions = []
-    objects = ndi.find_objects(label_image)
-    for i, sl in enumerate(objects, start=1):
-        if sl is None:
-            continue
-
-        regions.append(
-            acdcRegionProperties(
-                sl,
-                i,
-                label_image,
-                intensity_image,
-                cache,
-                spacing=spacing,
-                extra_properties=extra_properties,
+    if _CYTHON_FIND_OBJECTS:
+        img_uint32 = label_image.astype(np.uint32, copy=False)
+        if label_image.ndim == 2:
+            objects = find_all_objects_2D(img_uint32)
+            for label, (r0, r1, c0, c1) in objects:
+                sl = (slice(r0, r1), slice(c0, c1))
+                regions.append(acdcRegionProperties(
+                    sl, label, label_image, intensity_image, cache,
+                    spacing=spacing, extra_properties=extra_properties,
+                    offset=offset_arr,
+                ))
+        else:
+            objects = find_all_objects_3D(img_uint32)
+            for label, (z0, z1, r0, r1, c0, c1) in objects:
+                sl = (slice(z0, z1), slice(r0, r1), slice(c0, c1))
+                regions.append(acdcRegionProperties(
+                    sl, label, label_image, intensity_image, cache,
+                    spacing=spacing, extra_properties=extra_properties,
+                    offset=offset_arr,
+                ))
+    else:
+        objects = ndi.find_objects(label_image)
+        for i, sl in enumerate(objects, start=1):
+            if sl is None:
+                continue
+            regions.append(acdcRegionProperties(
+                sl, i, label_image, intensity_image, cache,
+                spacing=spacing, extra_properties=extra_properties,
                 offset=offset_arr,
-            )
-            )
+            ))
     return regions
 
 

cellacdc/regionprops.py

@@ -2,12 +2,13 @@
 # cython: boundscheck=False, wraparound=False, cdivision=True
 import numpy as np
 cimport numpy as np
-from libc.limits cimport INT_MAX
+from libc.limits cimport UINT_MAX
 
-def find_all_objects_2D(np.int32_t[:, :] label_img):
-    cdef int n_rows = label_img.shape[0]
-    cdef int n_cols = label_img.shape[1]
-    cdef int i, j, label, max_label = 0
+def find_all_objects_2D(np.uint32_t[:, :] label_img):
+    cdef Py_ssize_t n_rows = label_img.shape[0]
+    cdef Py_ssize_t n_cols = label_img.shape[1]
+    cdef Py_ssize_t i, j
+    cdef unsigned int label, max_label = 0
 
     # First pass: find max label to allocate C arrays
     for i in range(n_rows):
@@ -19,34 +20,35 @@ def find_all_objects_2D(np.int32_t[:, :] label_img):
     if max_label == 0:
         return []
 
-    cdef np.ndarray[np.int32_t, ndim=1] _rs = np.full(max_label + 1, INT_MAX, dtype=np.int32)
-    cdef np.ndarray[np.int32_t, ndim=1] _re = np.full(max_label + 1, -1,      dtype=np.int32)
-    cdef np.ndarray[np.int32_t, ndim=1] _cs = np.full(max_label + 1, INT_MAX, dtype=np.int32)
-    cdef np.ndarray[np.int32_t, ndim=1] _ce = np.full(max_label + 1, -1,      dtype=np.int32)
+    cdef np.ndarray[np.uint32_t, ndim=1] _rs = np.full(max_label + 1, UINT_MAX, dtype=np.uint32)
+    cdef np.ndarray[np.uint32_t, ndim=1] _re = np.zeros(max_label + 1,          dtype=np.uint32)
+    cdef np.ndarray[np.uint32_t, ndim=1] _cs = np.full(max_label + 1, UINT_MAX, dtype=np.uint32)
+    cdef np.ndarray[np.uint32_t, ndim=1] _ce = np.zeros(max_label + 1,          dtype=np.uint32)
 
-    cdef int[:] rs = _rs, re = _re, cs = _cs, ce = _ce
+    cdef unsigned int[:] rs = _rs, re = _re, cs = _cs, ce = _ce
 
     # Second pass: compute bounding boxes without Python objects in the hot loop
     for i in range(n_rows):
         for j in range(n_cols):
             label = label_img[i, j]
             if label > 0:
-                if i     < rs[label]: rs[label] = i
-                if i + 1 > re[label]: re[label] = i + 1
-                if j     < cs[label]: cs[label] = j
-                if j + 1 > ce[label]: ce[label] = j + 1
+                if i     < rs[label]: rs[label] = <unsigned int>i
+                if i + 1 > re[label]: re[label] = <unsigned int>(i + 1)
+                if j     < cs[label]: cs[label] = <unsigned int>j
+                if j + 1 > ce[label]: ce[label] = <unsigned int>(j + 1)
 
     return [
         (lbl, (rs[lbl], re[lbl], cs[lbl], ce[lbl]))
         for lbl in range(1, max_label + 1)
-        if re[lbl] != -1
+        if re[lbl] != 0
     ]
 
-def find_all_objects_3D(np.int32_t[:, :, :] label_img):
-    cdef int n_z = label_img.shape[0]
-    cdef int n_rows = label_img.shape[1]
-    cdef int n_cols = label_img.shape[2]
-    cdef int i, j, k, label, max_label = 0
+def find_all_objects_3D(np.uint32_t[:, :, :] label_img):
+    cdef Py_ssize_t n_z = label_img.shape[0]
+    cdef Py_ssize_t n_rows = label_img.shape[1]
+    cdef Py_ssize_t n_cols = label_img.shape[2]
+    cdef Py_ssize_t i, j, k
+    cdef unsigned int label, max_label = 0
 
     # First pass: find max label
     for i in range(n_z):
@@ -59,30 +61,30 @@ def find_all_objects_3D(np.int32_t[:, :, :] label_img):
     if max_label == 0:
         return []
 
-    cdef np.ndarray[np.int32_t, ndim=1] _zs = np.full(max_label + 1, INT_MAX, dtype=np.int32)
-    cdef np.ndarray[np.int32_t, ndim=1] _ze = np.full(max_label + 1, -1,      dtype=np.int32)
-    cdef np.ndarray[np.int32_t, ndim=1] _rs = np.full(max_label + 1, INT_MAX, dtype=np.int32)
-    cdef np.ndarray[np.int32_t, ndim=1] _re = np.full(max_label + 1, -1,      dtype=np.int32)
-    cdef np.ndarray[np.int32_t, ndim=1] _cs = np.full(max_label + 1, INT_MAX, dtype=np.int32)
-    cdef np.ndarray[np.int32_t, ndim=1] _ce = np.full(max_label + 1, -1,      dtype=np.int32)
+    cdef np.ndarray[np.uint32_t, ndim=1] _zs = np.full(max_label + 1, UINT_MAX, dtype=np.uint32)
+    cdef np.ndarray[np.uint32_t, ndim=1] _ze = np.zeros(max_label + 1,          dtype=np.uint32)
+    cdef np.ndarray[np.uint32_t, ndim=1] _rs = np.full(max_label + 1, UINT_MAX, dtype=np.uint32)
+    cdef np.ndarray[np.uint32_t, ndim=1] _re = np.zeros(max_label + 1,          dtype=np.uint32)
+    cdef np.ndarray[np.uint32_t, ndim=1] _cs = np.full(max_label + 1, UINT_MAX, dtype=np.uint32)
+    cdef np.ndarray[np.uint32_t, ndim=1] _ce = np.zeros(max_label + 1,          dtype=np.uint32)
 
-    cdef int[:] zs = _zs, ze = _ze, rs = _rs, re = _re, cs = _cs, ce = _ce
+    cdef unsigned int[:] zs = _zs, ze = _ze, rs = _rs, re = _re, cs = _cs, ce = _ce
 
     # Second pass: compute bounding boxes
     for i in range(n_z):
         for j in range(n_rows):
             for k in range(n_cols):
                 label = label_img[i, j, k]
                 if label > 0:
-                    if i     < zs[label]: zs[label] = i
-                    if i + 1 > ze[label]: ze[label] = i + 1
-                    if j     < rs[label]: rs[label] = j
-                    if j + 1 > re[label]: re[label] = j + 1
-                    if k     < cs[label]: cs[label] = k
-                    if k + 1 > ce[label]: ce[label] = k + 1
+                    if i     < zs[label]: zs[label] = <unsigned int>i
+                    if i + 1 > ze[label]: ze[label] = <unsigned int>(i + 1)
+                    if j     < rs[label]: rs[label] = <unsigned int>j
+                    if j + 1 > re[label]: re[label] = <unsigned int>(j + 1)
+                    if k     < cs[label]: cs[label] = <unsigned int>k
+                    if k + 1 > ce[label]: ce[label] = <unsigned int>(k + 1)
 
     return [
         (lbl, (zs[lbl], ze[lbl], rs[lbl], re[lbl], cs[lbl], ce[lbl]))
         for lbl in range(1, max_label + 1)
-        if ze[lbl] != -1
+        if ze[lbl] != 0
     ]