Local bundle update

.gitignore

@@ -16,3 +16,6 @@ data/berlin/*
 !data/berlin/images
 
 eval
+
+# Ignore virtualenv files
+env/*

README.md

@@ -30,7 +30,6 @@ Install OpenCV using
     brew install opencv
     brew install homebrew/science/ceres-solver
     brew install boost-python
-    sudo pip install -r requirements.txt
 
 And install OpenGV using
 
@@ -63,6 +62,10 @@ When running OpenSfM on top of [OpenCV][] 3.0 the [OpenCV Contrib][] modules are
 
 ## Building
 
+    sudo pip install virtualenv
+    virtualenv env
+    source env/bin/activate
+    pip install -r requirements.txt
     python setup.py build
 
 
@@ -72,7 +75,7 @@ An example dataset is available at `data/berlin`.
 
  1. Put some images in `data/DATASET_NAME/images/`
  2. Put config.yaml in `data/DATASET_NAME/config.yaml`
- 3. Go to the root of the project and run `bin/run_all data/DATASET_NAME`
+ 3. Go to the root of the project and run `bin/opensfm_run_all data/DATASET_NAME`
  4. Start an http server from the root with `python -m SimpleHTTPServer`
  5. Browse `http://localhost:8000/viewer/reconstruction.html#file=/data/DATASET_NAME/reconstruction.meshed.json`.
 

bin/plot_depthmaps

@@ -0,0 +1,76 @@
+#!/usr/bin/env python
+
+import argparse
+import matplotlib.pyplot as pl
+import numpy as np
+from textwrap import wrap
+
+from opensfm import dataset
+from opensfm import io
+
+
+def plot_depthmap(im, title, depth, plane, score, nghbr):
+        ax = pl.subplot2grid((2, 3), (0, 0), rowspan=2)
+        ax_title = ax.set_title(title)
+        ax_title.set_y(1.05)
+        pl.imshow(im)
+        ax = pl.subplot(2, 3, 2)
+        ax_title = ax.set_title("Depth")
+        ax_title.set_y(1.08)
+        pl.imshow(depth)
+        pl.colorbar()
+        ax = pl.subplot(2, 3, 3)
+        ax_title = ax.set_title("Score")
+        ax_title.set_y(1.08)
+        pl.imshow(score)
+        pl.colorbar()
+        ax = pl.subplot(2, 3, 5)
+        ax_title = ax.set_title("Neighbor")
+        ax_title.set_y(1.08)
+        pl.imshow(nghbr)
+        pl.colorbar()
+        ax = pl.subplot(2, 3, 6)
+        ax_title = ax.set_title("Plane normal")
+        ax_title.set_y(1.02)
+        pl.imshow(color_plane_normals(plane))
+
+
+def color_plane_normals(plane):
+    l = np.linalg.norm(plane, axis=2)
+    normal = plane / l[..., np.newaxis]
+    normal[..., 1] *= -1  # Reverse Y because it points down
+    normal[..., 2] *= -1  # Reverse Z because standard colormap does so
+    return ((normal + 1) * 128).astype(np.uint8)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description='Compute reconstruction')
+    parser.add_argument('dataset',
+                        help='path to the dataset to be processed')
+    parser.add_argument('--image',
+                        help='name of the image to show')
+    parser.add_argument('--save-figs',
+                        help='save figures istead of showing them',
+                        action='store_true')
+    args = parser.parse_args()
+
+    data = dataset.DataSet(args.dataset)
+
+    images = [args.image] if args.image else data.images()
+    for image in images:
+            depth, plane, score, nghbr, nghbrs = data.load_raw_depthmap(image)
+
+            print "Plotting depthmap for {0}".format(image)
+            pl.close("all")
+            pl.figure(figsize=(30, 16), dpi=90, facecolor='w', edgecolor='k')
+            title = "Shot: " + image + "\n" + "\n".join(wrap("Neighbors: " + ', '.join(nghbrs), 80))
+            plot_depthmap(data.image_as_array(image), title, depth, plane, score, nghbr)
+            pl.tight_layout()
+
+            if args.save_figs:
+                p = args.dataset + '/plot_depthmaps'
+                io.mkdir_p(p)
+                pl.savefig(p + '/' + image + '.png')
+                pl.close()
+            else:
+                pl.show()

bin/plot_matches

@@ -1,6 +1,7 @@
 #!/usr/bin/env python
 
 import argparse
+from itertools import combinations
 
 import matplotlib.pyplot as pl
 import networkx as nx
@@ -33,7 +34,9 @@ if __name__ == "__main__":
     parser.add_argument('dataset',
                         help='path to the dataset to be processed')
     parser.add_argument('--image',
-                        help='name of the image to show')
+                        help='show tracks for a specific')
+    parser.add_argument('--images',
+                        help='show tracks between a subset of images (separated by commas)')
     parser.add_argument('--graph',
                         help='display image graph',
                         action='store_true')
@@ -65,39 +68,40 @@ if __name__ == "__main__":
     else:
         # Plot matches between images
         if args.image:
-            toplot = [args.image]
+            pairs = [(args.image, o) for o in images if o != args.image]
+        elif args.images:
+            subset = args.images.split(',')
+            pairs = combinations(subset, 2)
         else:
-            toplot = images
+            pairs = combinations(images, 2)
 
         i = 0
-        for im1 in toplot:
-            for im2 in images:
-                if im1 != im2:
-                    matches = data.find_matches(im1, im2)
-                    if len(matches) == 0:
-                        continue
-                    print 'plotting matches between', im1, im2
-
-                    p1, f1, c1 = data.load_features(im1)
-                    p2, f2, c2 = data.load_features(im2)
-                    p1 = p1[matches[:, 0]]
-                    p2 = p2[matches[:, 1]]
-
-                    pl.figure(figsize=(20, 10))
-                    pl.title('Images: ' + im1 + ' - ' + im2 + ', matches: ' +
-                             str(matches.shape[0]))
-                    plot_matches(data.image_as_array(im1),
-                                 data.image_as_array(im2), p1, p2)
-                    i += 1
-                    if args.save_figs:
-                        p = args.dataset + '/plot_tracks'
-                        io.mkdir_p(p)
-                        pl.savefig(p + '/' + im1 + '_' + im2 + '.jpg', dpi=100)
-                        pl.close()
-                    else:
-                        if i >= 10:
-                            i = 0
-                            pl.show()
+        for im1, im2 in pairs:
+            matches = data.find_matches(im1, im2)
+            if len(matches) == 0:
+                continue
+            print 'plotting matches between', im1, im2
+
+            p1, f1, c1 = data.load_features(im1)
+            p2, f2, c2 = data.load_features(im2)
+            p1 = p1[matches[:, 0]]
+            p2 = p2[matches[:, 1]]
+
+            pl.figure(figsize=(20, 10))
+            pl.title('Images: ' + im1 + ' - ' + im2 + ', matches: ' +
+                     str(matches.shape[0]))
+            plot_matches(data.image_as_array(im1),
+                         data.image_as_array(im2), p1, p2)
+            i += 1
+            if args.save_figs:
+                p = args.dataset + '/plot_tracks'
+                io.mkdir_p(p)
+                pl.savefig(p + '/' + im1 + '_' + im2 + '.jpg', dpi=100)
+                pl.close()
+            else:
+                if i >= 10:
+                    i = 0
+                    pl.show()
 
         if not args.save_figs and i > 0:
             pl.show()

bin/plot_tracks

@@ -1,6 +1,8 @@
 #!/usr/bin/env python
 
 import argparse
+from itertools import combinations
+
 import matplotlib.pyplot as pl
 import networkx as nx
 import numpy as np
@@ -30,11 +32,13 @@ def plot_matches(im1, im2, p1, p2):
 
 
 if __name__ == "__main__":
-    parser = argparse.ArgumentParser(description='Compute reconstruction')
+    parser = argparse.ArgumentParser(description='Plot tracks')
     parser.add_argument('dataset',
                         help='path to the dataset to be processed')
     parser.add_argument('--image',
-                        help='name of the image to show')
+                        help='show tracks for a specific')
+    parser.add_argument('--images',
+                        help='show tracks between a subset of images (separated by commas)')
     parser.add_argument('--graph',
                         help='display image graph',
                         action='store_true')
@@ -60,32 +64,33 @@ if __name__ == "__main__":
     else:
         # Plot matches between images
         if args.image:
-            toplot = [args.image]
+            pairs = [(args.image, o) for o in images if o != args.image]
+        elif args.images:
+            subset = args.images.split(',')
+            pairs = combinations(subset, 2)
         else:
-            toplot = images
+            pairs = combinations(images, 2)
 
         i = 0
-        for im1 in toplot:
-            for im2 in images:
-                if im1 != im2:
-                    t, p1, p2 = matching.common_tracks(graph, im1, im2)
-                    if len(t) >= 10:
-                        pl.figure(figsize=(20, 10))
-                        pl.title('Images: ' + im1 + ' - ' + im2 +
-                                 ', matches: ' + str(len(t)))
-                        plot_matches(data.image_as_array(im1),
-                                     data.image_as_array(im2), p1, p2)
-                        i += 1
-                        if args.save_figs:
-                            p = args.dataset + '/plot_tracks'
-                            io.mkdir_p(p)
-                            pl.savefig(p + '/' + im1 + '_' + im2 + '.jpg',
-                                       dpi=100)
-                            pl.close()
-                        else:
-                            if i >= 10:
-                                i = 0
-                                pl.show()
+        for im1, im2 in pairs:
+            t, p1, p2 = matching.common_tracks(graph, im1, im2)
+            if len(t) >= 10:
+                pl.figure(figsize=(20, 10))
+                pl.title('Images: ' + im1 + ' - ' + im2 +
+                         ', matches: ' + str(len(t)))
+                plot_matches(data.image_as_array(im1),
+                             data.image_as_array(im2), p1, p2)
+                i += 1
+                if args.save_figs:
+                    p = args.dataset + '/plot_tracks'
+                    io.mkdir_p(p)
+                    pl.savefig(p + '/' + im1 + '_' + im2 + '.jpg',
+                               dpi=100)
+                    pl.close()
+                else:
+                    if i >= 10:
+                        i = 0
+                        pl.show()
 
         if not args.save_figs and i > 0:
             pl.show()

bin/run_eval

@@ -17,7 +17,7 @@ from opensfm import types
 
 def run_dataset(run_root, name):
     folder = run_root + '/' + name
-    call(['bin/run_all', folder])
+    call(['bin/opensfm_run_all', folder])
 
     with open(run_root + '/index.html', 'a') as fout:
         s = '''

config.yaml

@@ -29,10 +29,6 @@ akaze_descriptor_channels: 3  # Number of feature channels (1,2,3)
 hahog_peak_threshold: 0.00001
 hahog_edge_threshold: 10
 
-# Masks for regions that will be ignored for feature extraction
-# List of bounding boxes specified as the ratio to image width and height
-# masks: [{top: 0.96, bottom: 1.0, left: 0.0, right: 0.15}, {top: 0.95, bottom: 1.0, left: 0, right: 0.05}]
-
 # Params for general matching
 lowes_ratio: 0.8              # Ratio test for matches
 preemptive_lowes_ratio: 0.6   # Ratio test for preemptive matches

data/berlin/config.yaml

@@ -5,3 +5,4 @@
 processes: 8                  # Number of threads to use
 depthmap_min_consistent_views: 2      # Min number of views that should reconstruct a point for it to be valid
 depthmap_save_debug_files: yes        # Save debug files with partial reconstruction results
+feature_process_size: 1024

doc/Makefile

@@ -58,7 +58,7 @@ html:
 	@echo "Build finished. The HTML pages are in $(BUILDDIR)/html."
 
 livehtml: html
-	sphinx-autobuild -b html $(ALLSPHINXOPTS) $(BUILDDIR)/html
+	sphinx-autobuild -b html -p 8001 $(ALLSPHINXOPTS) $(BUILDDIR)/html
 
 dirhtml:
 	$(SPHINXBUILD) -b dirhtml $(ALLSPHINXOPTS) $(BUILDDIR)/dirhtml

doc/source/_static/mathjax_conf.js

@@ -1,6 +1,6 @@
 MathJax.Hub.Config({
    "HTML-CSS": {
         availableFonts: ["TeX"],
-        scale: 90
+        scale: 100
    }
 });

doc/source/building.rst

@@ -70,3 +70,12 @@ When running OpenSfM on top of OpenCV version 3.0 the `OpenCV Contrib`_ modules
 .. _Ceres solver: http://ceres-solver.org/
 .. _Boost Python: http://www.boost.org/
 .. _Networkx: https://github.com/networkx/networkx
+
+
+Building the documentation
+--------------------------
+To build the documentation and browse it locally use
+::
+    cd doc
+    make livehtml
+and browse `http://localhost:8001/ <http://localhost:8001/>`_

doc/source/conf.py

@@ -54,17 +54,17 @@
 
 # General information about the project.
 project = u'OpenSfM'
-copyright = u'2016, Mapillary'
+copyright = u'2017, Mapillary'
 author = u'Mapillary'
 
 # The version info for the project you're documenting, acts as replacement for
 # |version| and |release|, also used in various other places throughout the
 # built documents.
 #
 # The short X.Y version.
-version = u'0.0'
+version = u'0.1'
 # The full version, including alpha/beta/rc tags.
-release = u'0.0.0'
+release = u'0.1.0'
 
 # The language for content autogenerated by Sphinx. Refer to documentation
 # for a list of supported languages.

doc/source/dataset.rst

@@ -4,4 +4,21 @@
 Dataset Structure
 =================
 
-TODO
+::
+
+   project/
+   ├── config.yaml
+   ├── images/
+   ├── masks/
+   ├── gcp_list.txt
+   ├── metadata/
+   ├── features/
+   ├── matches/
+   ├── tracks.tsv
+   ├── reconstruction.json
+   ├── reconstruction.meshed.json
+   ├── undistorted/
+   ├── undistorted_tracks.json
+   ├── undistorted_reconstruction.json
+   └── depthmaps/
+       └── merged.ply

doc/source/dense.rst

@@ -120,7 +120,7 @@ And the linear approximation around :math:`(x_0, y_0)` is
 Undistortion
 ------------
 
-The dense module assumes that images are taken with perspective projection and no radial distortion.  For perspective images, undistorted versions can be generated by taking into account the computed distortion parameters, :math:`k1` and :math: `k2`.
+The dense module assumes that images are taken with perspective projection and no radial distortion.  For perspective images, undistorted versions can be generated by taking into account the computed distortion parameters, :math:`k1` and :math:`k2`.
 
 Equirectangular images (360 panoramas) however can not be unwarped into a single persepective view.  We need to generate multiple perspective views to cover the field of view of a panorama.