def predict_label(self, feature):
knn = compute_knn(self._labeled_features,
lambda lf: self.dist(feature, lf.feature), self._k)
label_counts = {}
for lf in knn:
if not lf.label in label_counts.keys():
label_counts[lf.label] = 0
label_counts[lf.label] += 1
return max(label_counts.keys(), key=lambda k: label_counts[k])
def predict_label( self, feature ):
knn = compute_knn( self._labeled_features,
lambda lf: self.dist(feature, lf.feature ),
self._k )
label_counts = {}
for lf in knn:
if not lf.label in label_counts.keys():
label_counts[lf.label] = 0
label_counts[lf.label] += 1
return max(label_counts.keys(), key = lambda k: label_counts[k])
def main(args):
compared_image = cv.LoadImage( args.compared_image)
reference_image = cv.LoadImage( args.reference_image )
compared_featuremap = FeatureMap()
compared_featuremap.read_from_file( args.compared_features )
reference_featuremap = FeatureMap()
reference_featuremap.read_from_file( args.reference_features )
compare_window = ClickWindow( compared_image, args.compare_zoom_out)
reference_window = ReferenceWindow( reference_image, args.reference_zoom_out)
reference_window.move( compare_window.size()[0], 0 )
#nn_solver = pyflann.FLANN()
while(True):
cont = update_all_windows()
if not cont:
break
if compare_window.update_nn:
click_pt = compare_window.click_pt
closest_pt = min( compared_featuremap.get_feature_points(),
key = lambda pt: l2_dist(pt,click_pt) )
compared_feature = compared_featuremap.get_feature( closest_pt )
distance_map = {}
shape_map = {}
size_map = {}
for pt in reference_featuremap.get_feature_points():
distance_map[pt] = l2_dist( compared_feature,
reference_featuremap.get_feature(pt) )
shape_map[pt] = reference_featuremap.get_shape(pt)
size_map[pt] = reference_featuremap.get_size(pt)
knn = compute_knn( distance_map.keys(), lambda pt: distance_map[pt], 20 )
reference_window.set_knn( knn )
reference_window.set_distance_map( distance_map )
reference_window.set_shape_map( shape_map )
reference_window.set_size_map( size_map )
compare_window.update_nn = False
