adapt to Numpy 1.25 changes…

(esp. `np.array(...)` now not allowed on ragged arrays unless
 `dtype=object`, but then coercing sub-arrays to `object` as well)

src/eynollah/eynollah.py

@@ -117,6 +117,7 @@ from .utils.marginals import get_marginals
 from .utils.resize import resize_image
 from .utils.shm import share_ndarray
 from .utils import (
+    ensure_array,
     is_image_filename,
     boosting_headers_by_longshot_region_segmentation,
     crop_image_inside_box,
@@ -2475,8 +2476,8 @@ class Eynollah:
             self, contours_only_text_parent, contours_only_text_parent_h, boxes, textline_mask_tot):
 
         self.logger.debug("enter do_order_of_regions")
-        contours_only_text_parent = np.array(contours_only_text_parent)
-        contours_only_text_parent_h = np.array(contours_only_text_parent_h)
+        contours_only_text_parent = ensure_array(contours_only_text_parent)
+        contours_only_text_parent_h = ensure_array(contours_only_text_parent_h)
         boxes = np.array(boxes, dtype=int) # to be on the safe side
         c_boxes = np.stack((0.5 * boxes[:, 2:4].sum(axis=1),
                             0.5 * boxes[:, 0:2].sum(axis=1)))
@@ -3987,7 +3988,7 @@ class Eynollah:
         def filterfun(lis):
             if len(lis) == 0:
                 return []
-            return list(np.array(lis)[indices])
+            return list(ensure_array(lis)[indices])
 
         return (filterfun(contours_par),
                 filterfun(contours_textline),
@@ -4378,7 +4379,8 @@ class Eynollah:
             areas_cnt_text = np.array([cv2.contourArea(c) for c in contours_only_text_parent])
             areas_cnt_text = areas_cnt_text / float(areas_tot_text)
             #self.logger.info('areas_cnt_text %s', areas_cnt_text)
-            contours_only_text_parent = np.array(contours_only_text_parent)[areas_cnt_text > MIN_AREA_REGION]
+            contours_only_text_parent = ensure_array(contours_only_text_parent)
+            contours_only_text_parent = contours_only_text_parent[areas_cnt_text > MIN_AREA_REGION]
             areas_cnt_text_parent = areas_cnt_text[areas_cnt_text > MIN_AREA_REGION]
 
             index_con_parents = np.argsort(areas_cnt_text_parent)
@@ -4397,12 +4399,13 @@ class Eynollah:
                 areas_cnt_text_d = np.array([cv2.contourArea(c) for c in contours_only_text_parent_d])
                 areas_cnt_text_d = areas_cnt_text_d / float(areas_tot_text_d)
 
-                contours_only_text_parent_d = np.array(contours_only_text_parent_d)[areas_cnt_text_d > MIN_AREA_REGION]
+                contours_only_text_parent_d = ensure_array(contours_only_text_parent_d)
+                contours_only_text_parent_d = contours_only_text_parent_d[areas_cnt_text_d > MIN_AREA_REGION]
                 areas_cnt_text_d = areas_cnt_text_d[areas_cnt_text_d > MIN_AREA_REGION]
 
                 if len(contours_only_text_parent_d):
                     index_con_parents_d = np.argsort(areas_cnt_text_d)
-                    contours_only_text_parent_d = np.array(contours_only_text_parent_d)[index_con_parents_d]
+                    contours_only_text_parent_d = contours_only_text_parent_d[index_con_parents_d]
                     areas_cnt_text_d = areas_cnt_text_d[index_con_parents_d]
 
                     centers_d = np.stack(find_center_of_contours(contours_only_text_parent_d)) # [2, N]
@@ -4546,7 +4549,8 @@ class Eynollah:
         #print("text region early 3 in %.1fs", time.time() - t0)
         if self.light_version:
             contours_only_text_parent = dilate_textregion_contours(contours_only_text_parent)
-            contours_only_text_parent , contours_only_text_parent_d_ordered = self.filter_contours_inside_a_bigger_one(
+            contours_only_text_parent, contours_only_text_parent_d_ordered = \
+                self.filter_contours_inside_a_bigger_one(
                     contours_only_text_parent, contours_only_text_parent_d_ordered, text_only,
                     marginal_cnts=polygons_of_marginals)
             #print("text region early 3.5 in %.1fs", time.time() - t0)

src/eynollah/utils/__init__.py

@@ -1,4 +1,4 @@
-from typing import List, Tuple
+from typing import Iterable, List, Tuple
 from logging import getLogger
 import time
 import math
@@ -1929,3 +1929,11 @@ def is_image_filename(fname: str) -> bool:
 
 def is_xml_filename(fname: str) -> bool:
     return fname.lower().endswith('.xml')
+
+def ensure_array(obj: Iterable) -> np.ndarray:
+    """convert sequence to array of type `object` so items can be of heterogeneous shape
+    (but ensure not to convert inner arrays to `object` if len=1)
+    """
+    if not isinstance(obj, np.ndarray):
+        return np.fromiter(obj, object)
+    return obj

src/eynollah/utils/contour.py

@@ -12,6 +12,7 @@ from shapely import set_precision
 from shapely.ops import unary_union, nearest_points
 
 from .rotate import rotate_image, rotation_image_new
+from . import ensure_array
 
 def contours_in_same_horizon(cy_main_hor):
     """
@@ -248,13 +249,15 @@ def return_contours_of_image(image):
     return contours, hierarchy
 
 def dilate_textline_contours(all_found_textline_polygons):
-    return [[polygon2contour(contour2polygon(contour, dilate=6))
-             for contour in region]
+    return [ensure_array(
+        [polygon2contour(contour2polygon(contour, dilate=6))
+         for contour in region])
             for region in all_found_textline_polygons]
 
-def dilate_textregion_contours(all_found_textline_polygons):
-    return [polygon2contour(contour2polygon(contour, dilate=6))
-            for contour in all_found_textline_polygons]
+def dilate_textregion_contours(all_found_textregion_polygons):
+    return ensure_array(
+        [polygon2contour(contour2polygon(contour, dilate=6))
+         for contour in all_found_textregion_polygons])
 
 def contour2polygon(contour: Union[np.ndarray, Sequence[Sequence[Sequence[Number]]]], dilate=0):
     polygon = Polygon([point[0] for point in contour])