def run(name):
inliers_threshold = 50.
n_samples = 5000
epsilon = 0
print(name)
path = '../data/{0}/'.format(name)
data = load(path)
model_class = fundamental.Fundamental
img_size = data['img2'].shape[:2]
nfa_proba = (2. * np.linalg.norm(img_size) / np.prod(img_size))
sampler = sampling.UniformSampler(n_samples)
generator = sampling.ModelGenerator(model_class, data['data'], sampler)
# generator = multigs.ModelGenerator(model_class, data['data'], n_samples)
min_sample_size = model_class().min_sample_size
ac_tester = ac.ImageTransformNFA(epsilon, nfa_proba, min_sample_size)
thresholder = membership.GlobalThresholder(inliers_threshold)
seed = 0
# seed = np.random.randint(0, np.iinfo(np.uint32).max)
print('seed:', seed)
np.random.seed(seed)
prefix = name
test_transformations.test(model_class, data, prefix, generator,
thresholder, ac_tester, name)
plt.close('all')
def run(subsampling=1, inliers_threshold=0.1, run_regular=True):
log_filename = 'logs/piazza_bra_s{0}.txt'.format(subsampling)
logger = utils.Logger(log_filename)
sys.stdout = logger
sigma = 1
epsilon = 0
local_ratio = 3.
name = 'Piazza_Bra'
dirname = '../data/' + name + '/'
mat = scipy.io.loadmat(dirname + 'Samantha_Bra.mat')
data = mat['Points']
# subsample the input points
points_considered = np.arange(0, data.shape[0], subsampling)
data = data[points_considered, :]
n_samples = data.shape[0] * 2
sampler = sampling.GaussianLocalSampler(sigma, n_samples)
ransac_gen = sampling.ModelGenerator(plane.Plane, data, sampler)
thresholder = membership.LocalThresholder(inliers_threshold,
ratio=local_ratio)
min_sample_size = plane.Plane().min_sample_size
ac_tester = ac.BinomialNFA(epsilon, 1. / local_ratio, min_sample_size)
seed = 0
# seed = np.random.randint(0, np.iinfo(np.uint32).max)
print('seed:', seed)
np.random.seed(seed)
output_prefix = name + '_n{0}'.format(data.shape[0])
test_3d.test(plane.Plane, data, output_prefix, ransac_gen, thresholder,
ac_tester, run_regular=run_regular)
plt.close('all')
sys.stdout = logger.stdout
logger.close()
return log_filename
def run(types, inliers_threshold=0.02, local_ratio=3., restimate_gt=False):
# Sampling ratio with respect to the number of elements
sampling_factor = 10
# a contrario test parameters
epsilon = 0.
config = {'Star5': line.Line,
'Star11': line.Line,
'Stairs': line.Line,
'Circles': circle.Circle,
}
stats_list = []
mat = scipy.io.loadmat('../data/JLinkageExamples.mat')
for example in mat.keys():
for c in types:
if example.find(c) == 0:
ex_type = c
break
else:
continue
model_class = config[ex_type]
data = mat[example].T
min_sample_size = model_class().min_sample_size
n_samples = data.shape[0] * sampling_factor * min_sample_size
sampler = sampling.UniformSampler(n_samples)
generator = sampling.ModelGenerator(model_class, data, sampler)
proba = 1. / local_ratio
ac_tester = ac.BinomialNFA(epsilon, proba, min_sample_size)
thresholder = membership.LocalThresholder(inliers_threshold,
ratio=local_ratio)
match = re.match('[a-zA-Z]+[0-9]*_', example)
try:
match = re.search('[0-9]+', match.group())
n_groups = int(match.group())
except AttributeError:
n_groups = 4
if restimate_gt:
gt_groups = ground_truth(data, n_groups, model_class=model_class,
thresholder=thresholder)
else:
gt_groups = ground_truth(data, n_groups)
seed = 0
# seed = np.random.randint(0, np.iinfo(np.uint32).max)
print('seed:', seed)
np.random.seed(seed)
output_prefix = example
if restimate_gt:
dir_name = 'test_2d_restimate_gt'
else:
dir_name = 'test_2d_given_gt'
res = test(model_class, data, output_prefix, generator, thresholder,
ac_tester, gt_groups, dir_name=dir_name)
stats_list.append(res)
print('-'*40)
plt.close('all')
reg_list, comp_list = zip(*stats_list)
print('Statistics of regular bi-clustering')
test_utils.compute_stats(reg_list)
print('Statistics of compressed bi-clustering')
test_utils.compute_stats(comp_list)
print('-'*40)
def evaluate_york(res_dir_name, run_with_lsd=False):
# RANSAC parameter
inliers_threshold = np.pi * 1e-2
logger = test_utils.Logger('logs/' + res_dir_name + '.txt')
sys.stdout = logger
dir_name = '/Users/mariano/Documents/datasets/YorkUrbanDB/'
sampling_factor = 4
epsilon = 0
local_ratio = 3.
ac_tester = ac.BinomialNFA(epsilon, 1. / local_ratio,
vp.VanishingPoint().min_sample_size)
thresholder = membership.LocalThresholder(inliers_threshold,
ratio=local_ratio)
stats_list = []
for i, example in enumerate(os.listdir(dir_name)):
if not os.path.isdir(dir_name + example):
continue
img_name = dir_name + '{0}/{0}.jpg'.format(example)
gray_image = PIL.Image.open(img_name).convert('L')
gt_name = dir_name + '{0}/{0}LinesAndVP.mat'.format(example)
mat = scipy.io.loadmat(gt_name)
gt_lines = mat['lines']
gt_segments = [lsd.Segment(gt_lines[k, :], gt_lines[k + 1, :])
for k in range(0, len(gt_lines), 2)]
gt_segments = np.array(gt_segments)
gt_association = np.squeeze(mat['vp_association'])
if run_with_lsd:
segments = lsd.compute(gray_image)
segments = np.array(segments)
gt_groups = ground_truth(gt_association, gt_segments, segments,
thresholder=thresholder)
else:
segments = gt_segments
gt_groups = [gt_association == v for v in np.unique(gt_association)]
sampler = sampling.UniformSampler(len(segments) * sampling_factor)
ransac_gen = sampling.ModelGenerator(vp.VanishingPoint, segments,
sampler)
seed = 0
# seed = np.random.randint(0, np.iinfo(np.uint32).max)
print('seed:', seed)
np.random.seed(seed)
res = test(gray_image, segments, res_dir_name, example, ransac_gen,
thresholder, ac_tester, gt_groups=gt_groups)
stats_list.append(res)
print('-'*40)
plt.close('all')
reg_list, comp_list = zip(*stats_list)
print('Statistics of regular bi-clustering')
test_utils.compute_stats(reg_list)
print('Statistics of compressed bi-clustering')
test_utils.compute_stats(comp_list)
sys.stdout = logger.stdout
logger.close()
def run(subsampling=1, inliers_threshold=0.1, run_regular=True):
log_filename = 'logs/pozzoveggiani_s{0}.txt'.format(subsampling)
logger = utils.Logger(log_filename)
sys.stdout = logger
sigma = 1
epsilon = 0
local_ratio = 3
name = 'PozzoVeggiani'
dirname = '../data/' + name + '/'
mat = scipy.io.loadmat(dirname + 'Results.mat')
data = mat['Points'].T
proj_mat = mat['Pmat']
visibility = mat['Visibility']
# Removing far away points for display
keep = functools.reduce(
np.logical_and,
[data[:, 0] > -10, data[:, 0] < 20, data[:, 2] > 10, data[:, 2] < 45])
data = data[keep, :]
visibility = visibility[keep, :]
# Re-order dimensions and invert vertical direction to get upright data
data[:, 1] *= -1
data = np.take(data, [0, 2, 1], axis=1)
proj_mat[:, 1, :] *= -1
proj_mat = np.take(proj_mat, [0, 2, 1, 3], axis=1)
# subsample the input points
points_considered = np.arange(0, data.shape[0], subsampling)
data = data[points_considered, :]
visibility = visibility[points_considered, :]
n_samples = data.shape[0] * 2
sampler = sampling.GaussianLocalSampler(sigma, n_samples)
generator = sampling.ModelGenerator(plane.Plane, data, sampler)
thresholder = membership.LocalThresholder(inliers_threshold,
ratio=local_ratio)
min_sample_size = plane.Plane().min_sample_size
ac_tester = ac.BinomialNFA(epsilon, 1. / local_ratio, min_sample_size)
projector = Projector(data, visibility, proj_mat, dirname)
seed = 0
# seed = np.random.randint(0, np.iinfo(np.uint32).max)
print('seed:', seed)
np.random.seed(seed)
output_prefix = name + '_n{0}'.format(data.shape[0])
test_3d.test(plane.Plane,
data,
output_prefix,
generator,
thresholder,
ac_tester,
plotter=projector,
run_regular=run_regular)
plt.close('all')
sys.stdout = logger.stdout
logger.close()
return log_filename