refactor: using opencv for faster convolution

detectree/image_descriptor.py

@@ -1,7 +1,7 @@
 """Compute image descriptors."""
+import cv2
 import numpy as np
 from PIL import Image
-from scipy import ndimage as ndi
 from skimage import color
 from skimage.util import shape
 from sklearn import preprocessing
@@ -60,7 +60,7 @@ def compute_image_descriptor(img_rgb, kernels, response_bins_per_axis, num_color
         )
 
     for i, kernel in enumerate(kernels):
-        filter_response = ndi.convolve(img_gray, kernel)
+        filter_response = cv2.filter2D(img_gray, ddepth=-1, kernel=kernel)
         response_bins = shape.view_as_blocks(filter_response, block_shape)
         bin_sum = response_bins.sum(axis=(2, 3)).flatten()
         gist_descr[i * num_blocks : (i + 1) * num_blocks] = bin_sum

detectree/pixel_features.py

@@ -1,8 +1,8 @@
 """Build pixel features."""
-
 import glob
 from os import path
 
+import cv2
 import dask
 import numpy as np
 from dask import diagnostics
@@ -180,7 +180,10 @@ def build_features_from_arr(self, img_rgb):
                 # theta = orientation / num_orientations * np.pi
                 theta = orientation * 180 / self.num_orientations
                 oriented_kernel_arr = ndi.interpolation.rotate(base_kernel_arr, theta)
-                img_filtered = ndi.convolve(img_lab_l, oriented_kernel_arr)
+                # img_filtered = ndi.convolve(img_lab_l, oriented_kernel_arr)
+                img_filtered = cv2.filter2D(
+                    img_lab_l, ddepth=-1, kernel=oriented_kernel_arr
+                )
                 img_filtered_vec = img_filtered.flatten()
                 X[
                     :, self.num_color_features + i * self.num_orientations + j

pyproject.toml

@@ -27,6 +27,7 @@ dependencies = [
     "joblib",
     "laspy >= 2.0.0",
     "numpy >= 1.15",
+    "opencv-python >= 4.0.0",
     "pandas >= 0.23",
     "pymaxflow >= 1.0.0",
     "rasterio >= 1.0.0",