Alternatives to opencv (cv2) SeldonIO#532 [Solved]

alibi_detect/utils/perturbation.py

@@ -2,7 +2,6 @@
 from io import BytesIO
 from typing import List, Tuple, Union
 
-import cv2
 import numpy as np
 import skimage as sk
 from alibi_detect.utils.data import Bunch
@@ -14,6 +13,14 @@
 from scipy.ndimage.interpolation import map_coordinates
 from skimage.filters import gaussian
 
+try: 
+    import cv2
+except:
+    cv2 = None
+    from scipy.signal import convolve2d
+    from skimage.transform import AffineTransform, warp
+    from skimage import img_as_ubyte
+
 
 def apply_mask(X: np.ndarray,
                mask_size: tuple = (4, 4),
@@ -632,8 +639,17 @@ def disk(radius: float, alias_blur: float = 0.1, dtype=np.float32) -> np.ndarray
     aliased_disk = np.array((X ** 2 + Y ** 2) <= radius ** 2, dtype=dtype)
     aliased_disk /= np.sum(aliased_disk)
 
-    # supersample disk to antialias
-    return cv2.GaussianBlur(aliased_disk, ksize=ksize, sigmaX=alias_blur)
+    if cv2 is not None:
+        # supersample disk to antialias
+        return cv2.GaussianBlur(aliased_disk, ksize=ksize, sigmaX=alias_blur)
+
+    else:
+        # Create Gaussian kernel
+        kernel = np.exp(-(X**2 + Y**2) / (2.0 * alias_blur**2))
+        kernel /= np.sum(kernel)
+
+        # Convolve with the aliased disk
+        return convolve2d(aliased_disk, kernel, mode='same', boundary='wrap')
 
 
 def defocus_blur(x: np.ndarray, radius: int, alias_blur: float, xrange: tuple = None) -> np.ndarray:
@@ -658,9 +674,16 @@ def defocus_blur(x: np.ndarray, radius: int, alias_blur: float, xrange: tuple =
     x, scale_back = scale_minmax(x, xrange)
     kernel = disk(radius=radius, alias_blur=alias_blur)
     channels = []
-    for d in range(3):
-        channels.append(cv2.filter2D(x[:, :, d], -1, kernel))
-    x_db = np.array(channels).transpose((1, 2, 0))
+
+    if cv2 is not None:
+        for d in range(3):
+            channels.append(cv2.filter2D(x[:, :, d], -1, kernel))
+            x_db = np.array(channels).transpose((1, 2, 0))
+    else:
+        for d in range(3):
+            channels.append(convolve2d(x[:, :, d], kernel, mode='same', boundary='wrap'))
+            x_db = np.array(channels).transpose((1, 2, 0)).astype(np.float32)
+
     if scale_back:
         x_db = x_db * (xrange[1] - xrange[0]) + xrange[0]
     if isinstance(xrange, tuple):
@@ -941,8 +964,19 @@ def elastic_transform(x: np.ndarray, mult_dxdy: float, sigma: float,
                        [center_square[0] + square_size, center_square[1] - square_size],
                        center_square - square_size], dtype=np.float32)
     pts2 = pts1 + np.random.uniform(-rnd_rng, rnd_rng, size=pts1.shape).astype(np.float32)
-    M = cv2.getAffineTransform(pts1, pts2)
-    image = cv2.warpAffine(x, M, shape_size[::-1], borderMode=cv2.BORDER_REFLECT_101)
+
+    if cv2 is not None:
+        M = cv2.getAffineTransform(pts1, pts2)
+        image = cv2.warpAffine(x, M, shape_size[::-1], borderMode=cv2.BORDER_REFLECT_101)
+
+    else:
+        M = AffineTransform()
+        M.estimate(pts1, pts2)
+
+        image = warp(x, M.inverse, output_shape=shape_size[::-1], mode='reflect')
+        # image = warp(x, M, output_shape=shape_size[::-1], mode='reflect')
+        image = img_as_ubyte(image)
+
     dx = (gaussian(np.random.uniform(-1, 1, size=shape_size),
                    sigma, mode='reflect', truncate=3) * mult_dxdy).astype(np.float32)
     dy = (gaussian(np.random.uniform(-1, 1, size=shape_size),

setup.py

@@ -25,12 +25,17 @@ def readme():
         "pykeops>=2.0.0, <2.2.0",
         "torch>=1.7.0, <1.14.0"
     ],
+    # https://github.com/SeldonIO/alibi-detect/issues/532
+    "opencv-python": [
+        "opencv-python>=3.2.0, <5.0.0"
+    ],
     "all": [
         "prophet>=1.1.0, <2.0.0",
         "tensorflow_probability>=0.8.0, <0.22.0",
         "tensorflow>=2.2.0, !=2.6.0, !=2.6.1, <2.14.0",  # https://github.com/SeldonIO/alibi-detect/issues/375 and 387
         "pykeops>=2.0.0, <2.2.0",
         "torch>=1.7.0, <1.14.0"
+        "opencv-python>=3.2.0, <5.0.0"
     ],
 }
 
@@ -53,7 +58,7 @@ def readme():
         "numpy>=1.16.2, <2.0.0",
         "pandas>=1.0.0, <3.0.0",
         "Pillow>=5.4.1, <11.0.0",
-        "opencv-python>=3.2.0, <5.0.0",
+        # "opencv-python>=3.2.0, <5.0.0", # Eradicated cv2 (opencv) requirement: https://github.com/SeldonIO/alibi-detect/issues/532
         "scipy>=1.3.0, <2.0.0",
         'scikit-image>=0.14.2, !=0.17.1, <0.22',  # https://github.com/SeldonIO/alibi/issues/215
         "scikit-learn>=0.20.2, <2.0.0",