Removed unnecessary rendering code.

The tool can now run in a plain console, with no UI.

README.md

@@ -12,13 +12,29 @@ Regular photos are not likely to work correctly, because regular cameras have pe
 
 More information is available in the [Wiki](/zlogic/cybervision/wiki).
 
+## How to use it
+
+Download a release .whl file for your platform from [releases](/zlogic/cybervision/releases).
+
+Install the .whl file by running:
+
+```sheell
+pip3 install <filename>.whl
+```
+
+Run cybervision:
+
+```shell
+python3 -m cybervision <img1.tif> <img2.tif> --output-file=<out.png> [--no-interpolate]
+```
+
 ## Python version
 
 Cybervision was rewritten in Python (with C extensions).
 
 Originally, it was a full all-in-one tool built based on Qt and using a different approach.
-For more details about the C++ version, see [Releases](releases).
-The source code is available in the [branch_qt_sift](tree/branch_qt_sift) branch.
+For more details about the C++ version, see [Releases](/zlogic/cybervision/releases).
+The source code is available in the [branch_qt_sift](/zlogic/cybervision/tree/branch_qt_sift) branch.
 
 The Python rewrite focuses on the primary goal - generating a 3D surface from an image stereopair;
 anything else (like a UI) can be added separately.
@@ -34,4 +50,5 @@ anything else (like a UI) can be added separately.
 ## External libraries
 
 * [fast](https://www.edwardrosten.com/work/fast.html) keypoint detector
+* [scipy](https://scipy.org) and [numpy](https://numpy.org) for point interpolation
 * [Pillow](https://python-pillow.org) image library

cybervision/__main__.py

@@ -13,6 +13,10 @@ def main():
     parser.add_argument('img1')
     parser.add_argument('img2')
     parser.add_argument('--resize-scale', type=float, default=1.0)
+    parser.add_argument('--interpolate', dest='interpolate', action='store_true')
+    parser.add_argument('--no-interpolate', dest='interpolate', action='store_false')
+    parser.set_defaults(interpolate=True)
+    parser.add_argument('--output-file', required=True)
     args = parser.parse_args()
 
     img1 = SEMImage(args.img1, args.resize_scale)
@@ -25,8 +29,11 @@ def main():
     except NoMatchesFound as err:
         sys.exit(err)
 
-    v = Visualiser(img1.img, img2.img, reconstructor.get_matches(), reconstructor.points3d)
-    v.show_results()
+    v = Visualiser(img1.img, img2.img, reconstructor.points3d)
+    if args.interpolate:
+        v.save_surface_image_interpolated(args.output_file)
+    else:
+        v.save_surface_image(args.output_file)
 
 
 if __name__ == '__main__':

cybervision/reconstruction.py

@@ -136,10 +136,9 @@ def reconstruct(self):
             raise NoMatchesFound('Failed to fit the model')
 
         matches_count = len(self.matches)
-        if not self.keep_intermediate_results:
-            self.matches = []
-            self.points1 = []
-            self.points2 = []
+        del(self.matches)
+        del(self.points1)
+        del(self.points2)
 
         time_completed_ransac = datetime.now()
         self.log.info(f'Completed RANSAC fitting in {time_completed_ransac-time_completed_matching}')
@@ -151,9 +150,8 @@ def reconstruct(self):
         self.points3d = self.create_surface()
         w1 = self.img1.width
         h1 = self.img1.height
-        if not self.keep_intermediate_results:
-            del(self.img1)
-            del(self.img2)
+        del(self.img1)
+        del(self.img2)
 
         time_completed_surface = datetime.now()
         self.log.info(f'Completed surface generation in {time_completed_surface-time_completed_ransac}')
@@ -180,12 +178,11 @@ def get_matches(self):
             matches.append((p1[0], p1[1], p2[0], p2[1], corr))
         return matches
 
-    def __init__(self, img1: Image, img2: Image, keep_intermediate_results=False):
+    def __init__(self, img1: Image, img2: Image):
         self.img1 = img1.convert('L')
         self.img2 = img2.convert('L')
         self.log = logging.getLogger("reconstructor")
         self.num_threads = os.cpu_count()
-        self.keep_intermediate_results = keep_intermediate_results
 
         # Tunable parameters
         self.fast_threshold = 15

cybervision/visualisation.py

@@ -1,11 +1,9 @@
 from PIL import Image, ImageDraw
 
 import math
-import numpy as np
+import numpy
 import scipy.interpolate
 import scipy.spatial
-import matplotlib.pyplot as plt
-import matplotlib.tri as mtri
 
 # Sunset from https://jiffyclub.github.io/palettable/cartocolors/sequential/
 COLORMAP = [
@@ -35,84 +33,7 @@ def map_color(self, p):
         b = int(c2[2]*ratio+c1[2]*(1-ratio))
         return (r, g, b)
 
-    def show_points(self):
-        size = (self.img1.width+self.img2.width, max(self.img1.height, self.img2.height))
-        composite = Image.new("RGBA", size, (255, 255, 255, 0))
-        composite.paste(self.img1)
-        composite.paste(self.img2, (self.img1.width, 0))
-        draw = ImageDraw.Draw(composite)
-        for m in self.matches:
-            point1 = (m[0], m[1])
-            point2 = (m[2]+self.img1.width, m[3])
-            draw.point(point1, fill=(255, 0, 0, 255))
-            draw.point(point2, fill=(255, 0, 0, 255))
-        composite.show()
-
-    def show_matches(self):
-        size = (self.img1.width+self.img2.width, max(self.img1.height, self.img2.height))
-        composite = Image.new("RGBA", size, (255, 255, 255, 0))
-        composite.paste(self.img1)
-        composite.paste(self.img2, (self.img1.width, 0))
-        draw = ImageDraw.Draw(composite)
-
-        for m in self.matches:
-            point1 = (m[0], m[1])
-            point2 = (m[2]+self.img1.width, m[3])
-            draw.line(point1 + point2, fill=(255, 0, 0, 255))
-        for m in self.matches:
-            draw.point(point1, fill=(0, 255, 0, 255))
-            draw.point(point2, fill=(0, 255, 0, 255))
-
-        composite.show()
-
-    def show_distances(self):
-        size = (self.img1.width+self.img2.width, max(self.img1.height, self.img2.height))
-        composite = Image.new("RGBA", size, (255, 255, 255, 0))
-        composite.paste(self.img1)
-        composite.paste(self.img2, (self.img1.width, 0))
-        draw = ImageDraw.Draw(composite)
-
-        for m in self.matches:
-            point11 = (m[0], m[1])
-            point12 = (m[2], m[3])
-            point21 = (m[0]+self.img1.width, m[1])
-            point22 = (m[2]+self.img1.width, m[3])
-            draw.line(point11 + point12, fill=(255, 0, 0, 255))
-            draw.line(point21 + point22, fill=(255, 0, 0, 255))
-            draw.point(point11, fill=(0, 255, 0, 255))
-            draw.point(point12, fill=(0, 255, 0, 255))
-            draw.point(point21, fill=(0, 255, 0, 255))
-            draw.point(point22, fill=(0, 255, 0, 255))
-
-        composite.show()
-
-    def show_surface_plot(self):
-        x_coords = np.linspace(0, self.img1.width, self.img1.width, endpoint=False)
-        y_coords = np.linspace(0, self.img1.height, self.img1.height, endpoint=False)
-        xx, yy = np.meshgrid(x_coords, y_coords)
-        xy_coords = [(p[0], self.img1.height-p[1]) for p in self.points3d]
-        z_values = [p[2] for p in self.points3d]
-        interp_grid = scipy.interpolate.griddata(xy_coords, z_values, (xx, yy), method='linear')
-
-        ax = plt.axes(projection='3d')
-        ax.plot_surface(xx, yy, interp_grid, rcount=100, ccount=100, shade=True, cmap='jet')
-        plt.show()
-
-    def show_surface_mesh(self):
-        # Too slow without resampling
-        x = [p[0] for p in self.points3d]
-        y = [self.img1.height-p[1] for p in self.points3d]
-        xy_points = [[p[0], self.img1.height-p[1]] for p in self.points3d]
-        z_values = [p[2] for p in self.points3d]
-
-        mesh = scipy.spatial.Delaunay(xy_points)
-        triang = mtri.Triangulation(x=x, y=y, triangles=mesh.vertices)
-
-        ax = plt.axes(projection='3d')
-        ax.plot_trisurf(triang, z_values, cmap='jet', shade=False)
-        plt.show()
-
-    def show_surface_image(self):
+    def save_surface_image(self, filename):
         surface = Image.new("RGBA", self.img1.size, (0, 0, 0, 0))
         min_z = min(self.points3d, key=lambda p: p[2])[2]
         max_z = max(self.points3d, key=lambda p: p[2])[2]
@@ -122,39 +43,29 @@ def show_surface_image(self):
             (r, g, b) = self.map_color(z)
             draw.point((p[0], p[1]), fill=(r, g, b, 255))
 
-        surface.show()
+        surface.save(filename)
 
-    def show_surface_image_interpolated(self):
+    def save_surface_image_interpolated(self, filename):
         xy_points = [(p[0], p[1]) for p in self.points3d]
-        x_coords = np.linspace(0, self.img1.width, self.img1.width, endpoint=False)
-        y_coords = np.linspace(0, self.img1.height, self.img1.height, endpoint=False)
-        xx, yy = np.meshgrid(x_coords, y_coords)
+        x_coords = numpy.linspace(0, self.img1.width, self.img1.width, endpoint=False)
+        y_coords = numpy.linspace(0, self.img1.height, self.img1.height, endpoint=False)
+        xx, yy = numpy.meshgrid(x_coords, y_coords)
         z_values = [p[2] for p in self.points3d]
         depth_values = scipy.interpolate.griddata(xy_points, z_values, (xx, yy), method='linear')
 
         surface = Image.new("RGBA", self.img1.size, (0, 0, 0, 0))
-        min_z = np.nanmin(depth_values)
-        max_z = np.nanmax(depth_values)
-        for (ix, iy), z in np.ndenumerate(depth_values.T):
+        min_z = numpy.nanmin(depth_values)
+        max_z = numpy.nanmax(depth_values)
+        for (ix, iy), z in numpy.ndenumerate(depth_values.T):
             if not math.isnan(z):
                 x = round(x_coords[ix])
                 y = round(y_coords[iy])
                 z = (z-min_z)/(max_z-min_z)
                 (r, g, b) = self.map_color(z)
                 surface.putpixel((x, y), (r, g, b, 255))
-        surface.show()
-
-    def show_results(self):
-        # self.show_points()
-        # self.show_matches()
-        # self.show_distances()
-        # self.show_surface_plot()
-        # self.show_surface_mesh()
-        self.show_surface_image()
-        self.show_surface_image_interpolated()
+        surface.save(filename)
 
-    def __init__(self, img1: Image, img2: Image, matches, points3d):
+    def __init__(self, img1: Image, img2: Image, points3d):
         self.img1 = img1
         self.img2 = img2
-        self.matches = matches
         self.points3d = points3d

requirements.txt

@@ -1,11 +1,3 @@
-cycler==0.11.0
-fonttools==4.32.0
-kiwisolver==1.4.2
-matplotlib==3.5.1
 numpy==1.22.3
-packaging==21.3
 Pillow==9.1.0
-pyparsing==3.0.8
-python-dateutil==2.8.2
 scipy==1.8.0
-six==1.16.0

setup.py

@@ -1,23 +1,20 @@
 import sys
 from setuptools import setup, Extension
+from glob import glob
 
 extra_compile_args = []
 sources = [
         'machine/cybervision.c',
-        'machine/correlation.c',
-        'machine/fast/fast_9.c',
-        'machine/fast/fast_10.c',
-        'machine/fast/fast_11.c', 
-        'machine/fast/fast_12.c',
-        'machine/fast/fast.c',
-        'machine/fast/nonmax.c'
+        'machine/correlation.c'
     ]
 
 if sys.platform in ['darwin', 'linux']:
     extra_compile_args.append('-pthread')
 elif sys.platform == 'win32':
     sources.append('machine/win32/pthread.c')
 
+sources = sources + glob('machine/fast/*.c')
+
 machine = Extension(
     'cybervision.machine',
     sources=sources,
@@ -51,7 +48,6 @@
     setup_requires=['wheel'],
     install_requires=[
         "Pillow>=9.1.0",
-        "matplotlib>=3.5.1",
         "scipy>=1.8.0",
         "numpy>=1.22.3"
     ]