def get_points_over_angles_and_label_statistics(
coordinates: np.array, labels: np.array) -> (np.array, np.array):
azimuth = np.arctan2(coordinates[:, 1], coordinates[:, 0]) # [-pi, pi]
azimuth = np.mod(np.add(azimuth, 2 * np.pi), 2 * np.pi) # [0, 2pi]
elevation = np.arccos(
np.true_divide(coordinates[:, 2], np.linalg.norm(coordinates,
2,
axis=1))) # [0, pi]
elevation = np.where(elevation > np.pi / 2, -np.mod(elevation, np.pi / 2),
elevation) # [- pi/2, pi/2]
labels = labels[:, 0]
counter_a = get_counts(
azimuth,
labels,
get_num_learning_labels(),
np.multiply(azimuth_bins, np.pi / 180),
)
counter_e = get_counts(
elevation,
labels,
get_num_learning_labels(),
np.multiply(elevation_bins, np.pi / 180),
lower=True,
)
return counter_a, counter_e
def get_points_over_distance_and_label_statistics(coordinates: np.array, labels: np.array) \
-> (np.array, np.array, np.array):
depth = np.linalg.norm(coordinates, 2, axis=1)
labels = labels[:, 0]
counter = get_counts(depth, labels, get_num_learning_labels(),
np.array(distance_bins))
return np.sum(counter, axis=0), np.sum(counter, axis=1), counter
def draw_points_per_label(ppl: np.array, save_dir: str = None):
fig, ax = plt.subplots()
plt.bar(np.arange(get_num_learning_labels()), ppl)
ax.set_xlabel('Class Label')
ax.set_xticks(np.arange(get_num_learning_labels()))
ax.set_xticklabels(get_names_learning()[0],
ha='right',
rotation=45,
rotation_mode='anchor')
plt.ylabel('Number of Points')
plt.yscale('log')
plt.tight_layout()
if save_dir is None:
plt.show()
else:
plt.savefig(save_dir)
plt.close()
def analyse(path, mode, save_dir):
sequence_folder = os.path.join(path, 'dataset', 'sequences')
# placeholder for overall stats
pplds = np.zeros([11, get_num_learning_labels(),
len(distance_bins)],
dtype=np.int64)
# iterate over all train/val sequences
for s in range(11):
if mode == 'compute':
print('\nSequence {}'.format(s))
point_dir = os.path.join(sequence_folder, '{:02d}'.format(s),
'velodyne')
label_dir = os.path.join(sequence_folder, '{:02d}'.format(s),
'labels')
if not os.path.exists(point_dir):
raise NotADirectoryError(
'Directory for points does not exits: {}'.format(
point_dir))
if not os.path.exists(label_dir):
raise NotADirectoryError(
'Directory for labels does not exits: {}'.format(
label_dir))
if save_dir is not None and not os.path.exists(save_dir):
os.makedirs(save_dir)
points_per_distance = np.zeros(len(distance_bins), dtype=np.int64)
points_per_label = np.zeros(get_num_learning_labels(),
dtype=np.int64)
points_per_label_distance = np.zeros(
[get_num_learning_labels(),
len(distance_bins)],
dtype=np.int64)
points_per_label_azimuth = np.zeros(
[get_num_learning_labels(),
len(azimuth_bins)], dtype=np.int64)
points_per_label_elevation = np.zeros(
[get_num_learning_labels(),
len(elevation_bins) + 1],
dtype=np.int64)
point_filenames = sorted(os.listdir(point_dir))
label_filenames = sorted(os.listdir(label_dir))
for c, (point_file, label_file) in enumerate(
zip(point_filenames, label_filenames)):
sys.stdout.write('\r{:4d} / {:4d}'.format(
c, len(point_filenames)))
point_filename = os.path.join(point_dir, point_file)
label_filename = os.path.join(label_dir, label_file)
if not os.path.splitext(point_file)[0] == os.path.splitext(
label_file)[0]:
raise ValueError(
'Point list file {} does not match label file {}'.
format(point_filename, label_filename))
sensor, intensity = load_data_points(point_filename)
semantics, instance = load_data_labels(label_filename)
semantics = map_learning(semantics)
ppd, ppl, ppld = get_distance_label_stats(sensor, semantics)
ppla, pple = get_angle_label_stats(sensor, semantics)
points_per_distance = np.add(points_per_distance, ppd)
points_per_label = np.add(points_per_label, ppl)
points_per_label_distance = np.add(points_per_label_distance,
ppld)
points_per_label_azimuth = np.add(points_per_label_azimuth,
ppla)
points_per_label_elevation = np.add(points_per_label_elevation,
pple)
# save the data
if save_dir is not None:
np.savetxt(os.path.join(
save_dir, '{:02d}_points_per_distance.csv'.format(s)),
points_per_distance,
delimiter=',')
np.savetxt(os.path.join(
save_dir, '{:02d}_points_per_label.csv'.format(s)),
points_per_label,
delimiter=',')
np.savetxt(os.path.join(
save_dir, '{:02d}_distance_label_matrix.csv'.format(s)),
points_per_label_distance,
delimiter=',')
np.savetxt(os.path.join(
save_dir, '{:02d}_azimuth_label_matrix.csv'.format(s)),
points_per_label_azimuth,
delimiter=',')
np.savetxt(os.path.join(
save_dir, '{:02d}_elevation_label_matrix.csv'.format(s)),
points_per_label_elevation,
delimiter=',')
elif mode == 'from_data':
points_per_distance = np.loadtxt(os.path.join(
path, '{:02d}_points_per_distance.csv'.format(s)),
delimiter=',')
points_per_label = np.loadtxt(os.path.join(
path, '{:02d}_points_per_label.csv'.format(s)),
delimiter=',')
points_per_label_distance = np.loadtxt(os.path.join(
path, '{:02d}_distance_label_matrix.csv'.format(s)),
delimiter=',')
points_per_label_azimuth = np.loadtxt(os.path.join(
path, '{:02d}_azimuth_label_matrix.csv'.format(s)),
delimiter=',')
points_per_label_elevation = np.loadtxt(os.path.join(
path, '{:02d}_elevation_label_matrix.csv'.format(s)),
delimiter=',')
else:
raise ValueError
pplds[s] = points_per_label_distance
# visualize the data
plots.draw_distance_and_label_matrix(
points_per_label_distance,
save_dir=os.path.join(save_dir,
'{:02d}_distance_label_matrix'.format(s))
if save_dir is not None else None)
plots.draw_points_per_distance(
points_per_distance,
save_dir=os.path.join(save_dir,
'{:02d}_points_per_distance.'.format(s))
if save_dir is not None else None)
plots.draw_points_per_label(
points_per_label,
save_dir=os.path.join(save_dir,
'{:02d}_points_per_label'.format(s))
if save_dir is not None else None)
plots.draw_azimuth_and_label_matrix(
points_per_label_azimuth,
save_dir=os.path.join(save_dir,
'{:02d}_azimuth_label_matrix'.format(s))
if save_dir is not None else None)
plots.draw_elevation_and_label_matrix(
points_per_label_elevation,
save_dir=os.path.join(save_dir,
'{:02d}_elevation_label_matrix'.format(s))
if save_dir is not None else None)
print('\n')
print('Finished all sequences. Starting total analysis...')
ppld = np.sum(pplds, axis=0) # total points per label distance
ppl = np.sum(ppld, axis=1) # total points per label
ppd = np.sum(ppld, axis=0) # total points per distance
ppds = np.sum(pplds, axis=1) # points per distance sequence
ppls = np.sum(pplds, axis=2) # points per label sequence
# save the data
if save_dir is not None:
np.savetxt(os.path.join(save_dir, 'total_distance_label_matrix.csv'),
ppld,
delimiter=',')
np.savetxt(os.path.join(save_dir, 'total_points_per_distance.csv'),
ppd,
delimiter=',')
np.savetxt(os.path.join(save_dir, 'total_points_per_label.csv'),
ppl,
delimiter=',')
np.savetxt(os.path.join(save_dir, 'sequence_distance_matrix.csv'),
ppds,
delimiter=',')
np.savetxt(os.path.join(save_dir, 'sequence_label_matrix.csv'),
ppls,
delimiter=',')
# visualize the data
plots.draw_distance_and_label_matrix(
ppld,
save_dir=os.path.join(save_dir, 'total_distance_label_matrix')
if save_dir is not None else None)
plots.draw_points_per_distance(
ppd,
save_dir=os.path.join(save_dir, 'total_points_per_distance')
if save_dir is not None else None)
plots.draw_points_per_label(
ppl,
save_dir=os.path.join(save_dir, 'total_points_per_label')
if save_dir is not None else None)
plots.draw_distance_and_sequence_matrix(
ppds,
save_dir=os.path.join(save_dir, 'sequence_distance_matrix')
if save_dir is not None else None)
plots.draw_label_and_sequence_matrix(
ppls,
save_dir=os.path.join(save_dir, 'sequence_label_matrix')
if save_dir is not None else None)
plots.draw_sequence_length(save_dir=os.path.join(path, 'sequence_length')
if save_dir is not None else None)
def analyse(path, mode, save_dir):
sequence_folder = os.path.join(path, "dataset", "sequences")
# placeholder for overall stats
pplds = np.zeros([11, get_num_learning_labels(),
len(distance_bins)],
dtype=np.int64)
# iterate over all train/val sequences
for s in range(11):
if mode == "compute":
print("\nSequence {seq}".format(seq=s))
point_dir = os.path.join(sequence_folder, "{0:02d}".format(s),
"velodyne")
label_dir = os.path.join(sequence_folder, "{0:02d}".format(s),
"labels")
if not os.path.exists(point_dir):
raise NotADirectoryError(
"Directory for points does not exits: {dir}".format(
dir=point_dir))
if not os.path.exists(label_dir):
raise NotADirectoryError(
"Directory for labels does not exits: {dir}".format(
dir=label_dir))
if save_dir is not None and not os.path.exists(save_dir):
os.makedirs(save_dir)
points_per_distance = np.zeros(len(distance_bins), dtype=np.int64)
points_per_label = np.zeros(get_num_learning_labels(),
dtype=np.int64)
points_per_label_distance = np.zeros(
[get_num_learning_labels(),
len(distance_bins)],
dtype=np.int64)
points_per_label_azimuth = np.zeros(
[get_num_learning_labels(),
len(azimuth_bins)], dtype=np.int64)
points_per_label_elevation = np.zeros(
[get_num_learning_labels(),
len(elevation_bins) + 1],
dtype=np.int64)
point_filenames = sorted(os.listdir(point_dir))
label_filenames = sorted(os.listdir(label_dir))
for c, (point_file, label_file) in enumerate(
zip(point_filenames, label_filenames)):
sys.stdout.write("\r{0:4d} / {1:4d}".format(
c, len(point_filenames)))
point_filename = os.path.join(point_dir, point_file)
label_filename = os.path.join(label_dir, label_file)
if (not os.path.splitext(point_file)[0]
== os.path.splitext(label_file)[0]):
raise ValueError(
"Point list file {pf} does not match label file {lf}".
format(pf=point_filename, lf=label_filename))
sensor, intensity = load_data_points(point_filename)
semantics, instance = load_data_labels(label_filename)
semantics = map_learning(semantics)
ppd, ppl, ppld = get_distance_label_stats(sensor, semantics)
ppla, pple = get_angle_label_stats(sensor, semantics)
points_per_distance = np.add(points_per_distance, ppd)
points_per_label = np.add(points_per_label, ppl)
points_per_label_distance = np.add(points_per_label_distance,
ppld)
points_per_label_azimuth = np.add(points_per_label_azimuth,
ppla)
points_per_label_elevation = np.add(points_per_label_elevation,
pple)
# save the data
if save_dir is not None:
np.savetxt(
os.path.join(save_dir,
"{0:02d}_points_per_distance.csv".format(s)),
points_per_distance,
delimiter=",",
)
np.savetxt(
os.path.join(save_dir,
"{0:02d}_points_per_label.csv".format(s)),
points_per_label,
delimiter=",",
)
np.savetxt(
os.path.join(
save_dir,
"{0:02d}_distance_label_matrix.csv".format(s)),
points_per_label_distance,
delimiter=",",
)
np.savetxt(
os.path.join(save_dir,
"{0:02d}_azimuth_label_matrix.csv".format(s)),
points_per_label_azimuth,
delimiter=",",
)
np.savetxt(
os.path.join(
save_dir,
"{0:02d}_elevation_label_matrix.csv".format(s)),
points_per_label_elevation,
delimiter=",",
)
elif mode == "from_data":
points_per_distance = np.loadtxt(
os.path.join(path,
"{0:02d}_points_per_distance.csv".format(s)),
delimiter=",",
)
points_per_label = np.loadtxt(
os.path.join(path, "{0:02d}_points_per_label.csv".format(s)),
delimiter=",",
)
points_per_label_distance = np.loadtxt(
os.path.join(path,
"{0:02d}_distance_label_matrix.csv".format(s)),
delimiter=",",
)
points_per_label_azimuth = np.loadtxt(
os.path.join(path,
"{0:02d}_azimuth_label_matrix.csv".format(s)),
delimiter=",",
)
points_per_label_elevation = np.loadtxt(
os.path.join(path,
"{0:02d}_elevation_label_matrix.csv".format(s)),
delimiter=",",
)
else:
raise ValueError
pplds[s] = points_per_label_distance
# visualize the data
plots.draw_distance_and_label_matrix(
points_per_label_distance,
save_dir=os.path.join(save_dir,
"{0:02d}_distance_label_matrix".format(s))
if save_dir is not None else None,
)
plots.draw_points_per_distance(
points_per_distance,
save_dir=os.path.join(save_dir,
"{0:02d}_points_per_distance.".format(s))
if save_dir is not None else None,
)
plots.draw_points_per_label(
points_per_label,
save_dir=os.path.join(save_dir,
"{0:02d}_points_per_label".format(s))
if save_dir is not None else None,
)
plots.draw_azimuth_and_label_matrix(
points_per_label_azimuth,
save_dir=os.path.join(save_dir,
"{0:02d}_azimuth_label_matrix".format(s))
if save_dir is not None else None,
)
plots.draw_elevation_and_label_matrix(
points_per_label_elevation,
save_dir=os.path.join(save_dir,
"{0:02d}_elevation_label_matrix".format(s))
if save_dir is not None else None,
)
print("\n")
print("Finished all sequences. Starting total analysis...")
ppld = np.sum(pplds, axis=0) # total points per label distance
ppl = np.sum(ppld, axis=1) # total points per label
ppd = np.sum(ppld, axis=0) # total points per distance
ppds = np.sum(pplds, axis=1) # points per distance sequence
ppls = np.sum(pplds, axis=2) # points per label sequence
# save the data
if save_dir is not None:
np.savetxt(
os.path.join(save_dir, "total_distance_label_matrix.csv"),
ppld,
delimiter=",",
)
np.savetxt(os.path.join(save_dir, "total_points_per_distance.csv"),
ppd,
delimiter=",")
np.savetxt(os.path.join(save_dir, "total_points_per_label.csv"),
ppl,
delimiter=",")
np.savetxt(os.path.join(save_dir, "sequence_distance_matrix.csv"),
ppds,
delimiter=",")
np.savetxt(os.path.join(save_dir, "sequence_label_matrix.csv"),
ppls,
delimiter=",")
# visualize the data
plots.draw_distance_and_label_matrix(
ppld,
save_dir=os.path.join(save_dir, "total_distance_label_matrix")
if save_dir is not None else None,
)
plots.draw_points_per_distance(
ppd,
save_dir=os.path.join(save_dir, "total_points_per_distance")
if save_dir is not None else None,
)
plots.draw_points_per_label(
ppl,
save_dir=os.path.join(save_dir, "total_points_per_label")
if save_dir is not None else None,
)
plots.draw_distance_and_sequence_matrix(
ppds,
save_dir=os.path.join(save_dir, "sequence_distance_matrix")
if save_dir is not None else None,
)
plots.draw_label_and_sequence_matrix(
ppls,
save_dir=os.path.join(save_dir, "sequence_label_matrix")
if save_dir is not None else None,
)
plots.draw_sequence_length(save_dir=os.path.join(path, "sequence_length")
if save_dir is not None else None)