def compute_path_all_keypoints():
"""
"""
import os
import itertools
r = 6
c = 0.4
f = trav.tf_grayhist
image_path = 'aerial01.jpg'
tdigenerator = trav.TraversabilityEstimator(tf=f, r=r)
image = trav.load_image(os.path.join('image', image_path))
t_matrix = tdigenerator.get_traversability_matrix(image)
keypoints_image = trav.load_image(os.path.join('keypoints-positive', image_path))
grid = trav.grid_list(image, r)
keypoints = graphmapx.get_keypoints(keypoints_image, grid)
router = graphmapx.RouteEstimator(c=c)
G = router.tm2graph(t_matrix)
output_path = 'output/paths-%s-%d-0%d-%s' % (f.__name__, r, 100*c, image_path.split('.')[0])
if not os.path.exists(output_path):
os.makedirs(output_path)
for counter, (s, t) in enumerate(itertools.combinations(keypoints, 2)):
path, found = router.route(G, s, t)
path_indexes = [int(v) for v in path]
path_image = trav.draw_path(image, path_indexes, grid, found=found)
trav.save_image(os.path.join(output_path, 'path-%d.jpg' % (counter+1)), [path_image])
def traversability_matrix_histogram_plot():
"""
"""
import numpy
import matplotlib
import matplotlib.pyplot as plt
import seaborn
seaborn.set_style("darkgrid")
matplotlib.rcParams.update({'font.size': 14})
image = trav.load_image('image/aerial01.jpg')
mapper = trav.TraversabilityEstimator(r=6)
matrix = mapper.get_traversability_matrix(image, normalize=True)
fig, ax = plt.subplots(1,1)
weights = numpy.ones_like(matrix.flatten())/float(len(matrix.flatten()))
n, _, _ = ax.hist(matrix.flatten(), bins=numpy.arange(0, 1 + 0.1, 0.1), weights=weights, facecolor='green', alpha=0.5)
ax.set_xticks(numpy.arange(0, 1.01, 0.1))
ax.set_xlim([0, 1])
ax.set_yticks(numpy.arange(0.05, max(n)+0.05, 0.05))
ax.set_yticklabels(["%.0f%%" % (100*x) for x in numpy.arange(0.05, max(n)+0.05, 0.05)])
ax.grid(True)
fig.tight_layout()
plt.show()
def compute_path_random_keypoints():
"""
"""
import random
r = 6
c = 0.4
f = trav.tf_grayhist
tdigenerator = trav.TraversabilityEstimator(tf=f, r=r)
image = trav.load_image('image/aerial01.jpg')
t_matrix = tdigenerator.get_traversability_matrix(image)
keypoints_image = trav.load_image('keypoints-positive/aerial01.jpg')
grid = trav.grid_list(image, r)
keypoints = graphmapx.get_keypoints(keypoints_image, grid)
router = graphmapx.RouteEstimator(c=c)
G = router.tm2graph(t_matrix)
[source, target] = [v for v in random.sample(keypoints, 2)]
path, found = router.route(G, source, target)
# graphmap.draw_graph(G, 'output/path-graph.pdf', path) # only graph-tool
path_indexes = [int(v) for v in path]
path_image = trav.draw_path(image, path_indexes, grid, found=found)
trav.show_image([path_image])
def compute_traversability_matrix():
"""
"""
image = trav.load_image('image/aerial01.jpg')
mapper = trav.TraversabilityEstimator(r=6)
matrix = mapper.get_traversability_matrix(image)
print(matrix)
def compute_traversability_image():
"""
"""
image = trav.load_image('image/aerial01.jpg')
mapper = trav.TraversabilityEstimator(r=6)
t_image = mapper.get_traversability_image(image)
trav.show_image([t_image])
def compare_with_ground_truth():
"""
"""
mapper = trav.TraversabilityEstimator(r=6)
image = trav.load_image('image/aerial01.jpg')
ground_truth = trav.load_image('ground-truth/aerial01.jpg')
t_image = mapper.get_traversability_image(image)
t_ground_truth = mapper.get_ground_truth(ground_truth)
print("Correlation:", mapper.error(t_image, t_ground_truth, 'corr'))
trav.show_image([t_image, t_ground_truth])
def traversability_image_correlation_plot():
"""
"""
import os
import numpy
import matplotlib.pyplot as plt
dt_path = 'image'
gt_path = 'ground-truth'
output_path = 'output'
if not os.path.exists(output_path):
os.makedirs(output_path)
measure = 'corr'
f_set = [trav.tf_grayhist, trav.tf_rgbhist, trav.tf_superpixels]
r_set = [6, 12]
dataset = list()
for (_, _, filenames) in os.walk(gt_path):
dataset.extend(filenames)
break
dataset.sort(key=str.lower)
data = dict()
data[measure] = dict()
for image_path in dataset:
image = trav.load_image(os.path.join(dt_path, image_path))
ground_truth = trav.load_image(os.path.join(gt_path, image_path))
for f in f_set:
if not f.__name__ in data[measure]:
data[measure][f.__name__] = dict()
for r in r_set:
print("Running %s %s %s %s" % (measure, image_path, f.__name__, str(r)))
if not str(r) in data[measure][f.__name__]:
data[measure][f.__name__][str(r)] = list()
mapper = trav.TraversabilityEstimator(tf=f, r=r)
t_image = mapper.get_traversability_matrix(image)
t_ground_truth = mapper.get_ground_truth(ground_truth, matrix=True)
data[measure][f.__name__][str(r)].append(mapper.error(t_image, t_ground_truth, measure))
plot_title = {
"corr" : "Pearson's correlation coefficient",
"jaccard" : "Generalized Jaccard similarity index",
"rmse" : "Root mean square error",
"nrmse" : "Normalized root mean square error",
"psnr" : "Peak signal to noise ratio",
"ssim" : "Structural similarity index",
"nmae" : "Normalized mean absolute error"
}
ft_curve = {
"tf_random" : "Random",
"tf_grayhist" : "Gray Histogram",
"tf_rgbhist" : "RGB Histogram",
"tf_superpixels" : "Superpixels"
}
fig, (ax0) = plt.subplots(ncols=1)
for f in f_set:
x = numpy.array(r_set)
y = numpy.array([numpy.mean(data[measure][f.__name__][str(element)]) for element in x])
ax0.plot(x, y, '-o', markevery=range(len(x)), label=ft_curve[f.__name__])
plt.title(plot_title[measure])
#ax0.legend(loc='upper left')
#ax0.set_xscale('log')
ax0.set_xlabel("r")
ax0.set_ylabel(plot_title[measure].split(" ")[-1].title())
ax0.set_xticks(r_set)
ax0.set_xticklabels(["%dx%d" % (r, r) for r in r_set])
ax0.tick_params(axis='x', which='minor', bottom='off')
fig.tight_layout()
fig.savefig(os.path.join(output_path, "score-traversability_image-%s.png" % (measure)), dpi=300, bbox_inches='tight')
plt.close(fig)
def main_experiment():
"""
"""
import os
import time
import json
# import tqdm
import numpy
import itertools
dataset_path = 'image/'
ground_truth_path = 'ground-truth/'
positive_keypoints_path = 'keypoints-positive/'
negative_keypoints_path = 'keypoints-negative/'
output_path = 'output/'
output_file = 'data.json'
if not os.path.exists(output_path):
os.makedirs(output_path)
images = ['aerial%02d.jpg' % i for i in [8]]
f_set = [trav.reference, trav.tf_grayhist, trav.tf_rgbhist, trav.tf_superpixels]
r_set = [6, 8, 10, 12, 14, 16, 18, 20, 22, 24]
c_set = [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0]
dataset = list()
for (_, _, filenames) in os.walk(ground_truth_path):
dataset.extend(filenames)
break
selected = list(set(dataset).intersection(images)) if len(images) > 0 else dataset
selected.sort()
if os.path.exists(os.path.join(output_path, output_file)):
with open(os.path.join(output_path, output_file)) as datafile:
data = json.load(datafile)
else:
data = list()
# for i in tqdm.trange(len(selected), desc=" Input image "):
for i in range(len(selected)):
image_path = selected[i]
image = trav.load_image(os.path.join(dataset_path, image_path))
ground_truth = trav.load_image(os.path.join(ground_truth_path, image_path))
positive_keypoints = trav.load_image(os.path.join(positive_keypoints_path, image_path))
negative_keypoints = trav.load_image(os.path.join(negative_keypoints_path, image_path))
# for j in tqdm.trange(len(f_set), desc="Traversability function "):
for j in range(len(f_set)):
f = f_set[j]
# for k in tqdm.trange(len(r_set), desc=" Region size "):
for k in range(len(r_set)):
r = r_set[k]
penalty = r*(0.2/6)
mapper = trav.TraversabilityEstimator(tf=f, r=r)
start_matrix_time = time.time()
t_matrix = mapper.get_traversability_matrix(image)
matrix_time = time.time() - start_matrix_time
gt_matrix = mapper.get_ground_truth(ground_truth, matrix=True)
grid = trav.grid_list(image, r)
# for ii in tqdm.trange(len(c_set), desc=" Cut threshold "):
for ii in range(len(c_set)):
c = c_set[ii]
print("Processing: %s, tf: %s, r=%d, c=%.1f" % (image_path, f.__name__, r, c))
router = graphmapx.RouteEstimator(c=c)
start_graph_time = time.time()
if f == trav.reference:
G = router.tm2graph(gt_matrix)
else:
G = router.tm2graph(t_matrix)
graph_time = time.time() - start_graph_time
keypoints = graphmapx.get_keypoints(positive_keypoints, grid)
combinations = list(itertools.combinations(keypoints, 2))
# for counter in tqdm.trange(len(combinations), desc=" Positive paths "):
for counter in range(len(combinations)):
(s, t) = combinations[counter]
score = 1.0
start_route_time = time.time()
path, found = router.route(G, s, t)
route_time = time.time() - start_route_time
path_region_coordinates = [trav.coord(int(v), gt_matrix.shape[1]) for v in path]
for row, column in path_region_coordinates:
if gt_matrix[row][column] < 0.20:
score = numpy.maximum(0, score - penalty)
results = dict()
results['image'] = image_path
results['traversability_function'] = f.__name__
results['region_size'] = r
results['cut_threshold'] = c
results['path_existence'] = True
results['matrix_build_time'] = matrix_time
results['graph_build_time'] = graph_time
results['path_build_time'] = route_time
results['path_found'] = found
results['path_score'] = score if found else 0.0
results['path_regions'] = path_region_coordinates
data.append(results)
keypoints = graphmapx.get_keypoints(negative_keypoints, grid)
combinations = list(itertools.combinations(keypoints, 2))
# for counter in tqdm.trange(len(combinations), desc=" Negative paths "):
for counter in range(len(combinations)):
(s, t) = combinations[counter]
start_route_time = time.time()
path, found = router.route(G, s, t)
route_time = time.time() - start_route_time
results = dict()
results['image'] = image_path
results['traversability_function'] = f.__name__
results['region_size'] = r
results['cut_threshold'] = c
results['matrix_build_time'] = matrix_time
results['graph_build_time'] = graph_time
results['path_build_time'] = route_time
results['path_existence'] = False
results['path_found'] = found
results['path_score'] = 1.0 if not found else 0.0
results['path_regions'] = path_region_coordinates
data.append(results)
with open(os.path.join(output_path, output_file), 'w') as datafile:
json.dump(data, datafile, indent=4)