def calc_euroc_seq_errors(est_traj, gt_traj):
gt_traj_synced, est_traj_synced = sync.associate_trajectories(
gt_traj, est_traj, max_diff=0.01)
est_traj_aligned = trajectory.align_trajectory(est_traj_synced,
gt_traj_synced,
correct_scale=False,
correct_only_scale=False)
pose_relation = metrics.PoseRelation.translation_part
ape_metric = metrics.APE(pose_relation)
ape_metric.process_data((
gt_traj_synced,
est_traj_aligned,
))
ape_stat = ape_metric.get_statistic(metrics.StatisticsType.rmse)
# ape_metric = metrics.RPE(pose_relation)
# ape_metric.process_data((gt_traj_synced, est_traj_aligned,))
# ape_stat = ape_metric.get_statistic(metrics.StatisticsType.rmse)
# fig = plt.figure()
# traj_by_label = {
# "estimate (not aligned)": est_traj,
# "estimate (aligned)": est_traj_aligned,
# "reference": gt_traj
# }
# plot.trajectories(fig, traj_by_label, plot.PlotMode.xyz)
# plt.show()
return ape_metric, ape_stat
def test_se3_alignment(self):
traj = helpers.fake_trajectory(1000, 1)
traj_transformed = copy.deepcopy(traj)
traj_transformed.transform(lie.random_se3())
self.assertNotEqual(traj, traj_transformed)
traj_aligned = trajectory.align_trajectory(traj_transformed, traj)
self.assertEqual(traj_aligned, traj)
def test_alignment_degenerate_case(self):
length = 100
poses = [lie.random_se3()] * length
traj_1 = PoseTrajectory3D(poses_se3=poses,
timestamps=helpers.fake_timestamps(
length, 1, 0.0))
traj_2 = copy.deepcopy(traj_1)
traj_2.transform(lie.random_se3())
traj_2.scale(1.234)
self.assertNotEqual(traj_1, traj_2)
with self.assertRaises(GeometryException):
trajectory.align_trajectory(traj_1, traj_2)
with self.assertRaises(GeometryException):
trajectory.align_trajectory(traj_1, traj_2, correct_scale=True)
def test_scale_correction(self):
traj = helpers.fake_trajectory(1000, 1)
traj_transformed = copy.deepcopy(traj)
traj_transformed.scale(1.234)
self.assertNotEqual(traj, traj_transformed)
traj_aligned = trajectory.align_trajectory(traj_transformed, traj,
correct_only_scale=True)
self.assertEqual(traj_aligned, traj)
def test_sim3_alignment(self):
traj = helpers.fake_trajectory(1000, 1)
traj_transformed = copy.deepcopy(traj)
traj_transformed.transform(lie.random_se3())
traj_transformed.scale(1.234)
self.assertNotEqual(traj, traj_transformed)
traj_aligned = trajectory.align_trajectory(traj_transformed, traj,
correct_scale=True)
self.assertEqual(traj_aligned, traj)
def rpe(traj_ref, traj_est, pose_relation, delta, delta_unit,
rel_delta_tol=0.1, all_pairs=False, align=False, correct_scale=False,
ref_name="reference", est_name="estimate", support_loop=False):
from evo.core import metrics
from evo.core import trajectory
# Align the trajectories.
only_scale = correct_scale and not align
if align or correct_scale:
logger.debug(SEP)
traj_est = trajectory.align_trajectory(traj_est, traj_ref,
correct_scale, only_scale)
# Calculate RPE.
logger.debug(SEP)
data = (traj_ref, traj_est)
rpe_metric = metrics.RPE(pose_relation, delta, delta_unit, rel_delta_tol,
all_pairs)
rpe_metric.process_data(data)
title = str(rpe_metric)
if align and not correct_scale:
title += "\n(with SE(3) Umeyama alignment)"
elif align and correct_scale:
title += "\n(with Sim(3) Umeyama alignment)"
elif only_scale:
title += "\n(scale corrected)"
else:
title += "\n(not aligned)"
rpe_result = rpe_metric.get_result(ref_name, est_name)
rpe_result.info["title"] = title
logger.debug(SEP)
logger.info(rpe_result.pretty_str())
# Restrict trajectories to delta ids for further processing steps.
if support_loop:
# Avoid overwriting if called repeatedly e.g. in Jupyter notebook.
import copy
traj_ref = copy.deepcopy(traj_ref)
traj_est = copy.deepcopy(traj_est)
traj_ref.reduce_to_ids(rpe_metric.delta_ids)
traj_est.reduce_to_ids(rpe_metric.delta_ids)
rpe_result.add_trajectory(ref_name, traj_ref)
rpe_result.add_trajectory(est_name, traj_est)
if isinstance(traj_est, trajectory.PoseTrajectory3D) and not all_pairs:
seconds_from_start = [
t - traj_est.timestamps[0] for t in traj_est.timestamps
]
rpe_result.add_np_array("seconds_from_start", seconds_from_start)
rpe_result.add_np_array("timestamps", traj_est.timestamps)
return rpe_result
def ape(traj_ref,
traj_est,
pose_relation,
align=False,
correct_scale=False,
align_origin=False,
ref_name="reference",
est_name="estimate"):
from evo.core import metrics
from evo.core import trajectory
# Align the trajectories.
only_scale = correct_scale and not align
if align or correct_scale:
logger.debug(SEP)
traj_est = trajectory.align_trajectory(traj_est, traj_ref,
correct_scale, only_scale)
elif align_origin:
logger.debug(SEP)
traj_est = trajectory.align_trajectory_origin(traj_est, traj_ref)
# Calculate APE.
logger.debug(SEP)
data = (traj_ref, traj_est)
ape_metric = metrics.APE(pose_relation)
ape_metric.process_data(data)
title = str(ape_metric)
if align and not correct_scale:
title += "\n(with SE(3) Umeyama alignment)"
elif align and correct_scale:
title += "\n(with Sim(3) Umeyama alignment)"
elif only_scale:
title += "\n(scale corrected)"
elif align_origin:
title += "\n(with origin alignment)"
else:
title += "\n(not aligned)"
ape_result = ape_metric.get_result(ref_name, est_name)
ape_result.info["title"] = title
logger.debug(SEP)
logger.info(ape_result.pretty_str())
ape_result.add_trajectory(ref_name, traj_ref)
ape_result.add_trajectory(est_name, traj_est)
if isinstance(traj_est, trajectory.PoseTrajectory3D):
seconds_from_start = [
t - traj_est.timestamps[0] for t in traj_est.timestamps
]
ape_result.add_np_array("seconds_from_start", seconds_from_start)
ape_result.add_np_array("timestamps", traj_est.timestamps)
return ape_result
def original_ape(traj_ref,
traj_est,
pose_relation,
align=False,
correct_scale=False,
align_origin=False,
ref_name="reference",
est_name="estimate"):
''' Copied from main_ape.py
'''
traj_ref, traj_est = sync.associate_trajectories(traj_ref, traj_est)
# Align the trajectories.
only_scale = correct_scale and not align
if align or correct_scale:
traj_est = trajectory.align_trajectory(traj_est, traj_ref,
correct_scale, only_scale)
elif align_origin:
traj_est = trajectory.align_trajectory_origin(traj_est, traj_ref)
# Calculate APE.
data = (traj_ref, traj_est)
ape_metric = metrics.APE(pose_relation)
ape_metric.process_data(data)
title = str(ape_metric)
if align and not correct_scale:
title += "\n(with SE(3) Umeyama alignment)"
elif align and correct_scale:
title += "\n(with Sim(3) Umeyama alignment)"
elif only_scale:
title += "\n(scale corrected)"
elif align_origin:
title += "\n(with origin alignment)"
else:
title += "\n(not aligned)"
ape_result = ape_metric.get_result(ref_name, est_name)
ape_result.info["title"] = title
ape_result.add_trajectory(ref_name, traj_ref)
ape_result.add_trajectory(est_name, traj_est)
if isinstance(traj_est, trajectory.PoseTrajectory3D):
seconds_from_start = [
t - traj_est.timestamps[0] for t in traj_est.timestamps
]
ape_result.add_np_array("seconds_from_start", seconds_from_start)
ape_result.add_np_array("timestamps", traj_est.timestamps)
return ape_result
def three_plots(ref, est, table, name):
"""Generates plots and statistics table into Report
:ref: PoseTrajectory3D object that is used as reference
:est: PoseTrajectory3D object that is plotted against reference
:table: Tabular object that is generated by Tabular('c c')
:name: String that is used as name for file and table entry
:returns: translation of reference against estimation
"""
ref, est = sync.associate_trajectories(ref, est)
est, rot, tra, s = trajectory.align_trajectory(est,
ref,
correct_scale=False,
return_parameters=True)
data = (ref, est)
ape_metric = metrics.APE(metrics.PoseRelation.translation_part)
ape_metric.process_data(data)
ape_statistics = ape_metric.get_all_statistics()
# [ Localization ]
fig, axarr = plt.subplots(3) #sharex=True)
fig.suptitle('Localization', fontsize=30)
fig.tight_layout()
plot.traj_xyyaw(axarr, est, '-', 'red', 'estimation', 1, ref.timestamps[0])
plot.traj_xyyaw(axarr, ref, '-', 'gray', 'original')
fig.subplots_adjust(hspace=0.2)
plt.waitforbuttonpress(0)
plt.savefig("/home/kostas/results/latest/" + name + ".png",
format='png',
bbox_inches='tight')
plt.close(fig)
table.add_row((
name,
round(ape_statistics["rmse"], 3),
round(ape_statistics["mean"], 3),
round(ape_statistics["median"], 3),
round(ape_statistics["std"], 3),
round(ape_statistics["min"], 3),
round(ape_statistics["max"], 3),
round(ape_statistics["sse"], 3),
))
table.add_hline
def ape(traj_ref, traj_est, pose_relation, align=False,
correct_scale=False, ref_name="reference", est_name="estimate"):
from evo.core import metrics
from evo.core import trajectory
# Align the trajectories.
only_scale = correct_scale and not align
if align or correct_scale:
logger.debug(SEP)
traj_est = trajectory.align_trajectory(
traj_est, traj_ref, correct_scale, only_scale)
# Calculate APE.
logger.debug(SEP)
data = (traj_ref, traj_est)
ape_metric = metrics.APE(pose_relation)
ape_metric.process_data(data)
title = str(ape_metric)
if align and not correct_scale:
title += "\n(with SE(3) Umeyama alignment)"
elif align and correct_scale:
title += "\n(with Sim(3) Umeyama alignment)"
elif only_scale:
title += "\n(scale corrected)"
else:
title += "\n(not aligned)"
ape_result = ape_metric.get_result(ref_name, est_name)
ape_result.info["title"] = title
logger.debug(SEP)
logger.info(ape_result.pretty_str())
ape_result.add_trajectory(ref_name, traj_ref)
ape_result.add_trajectory(est_name, traj_est)
if isinstance(traj_est, trajectory.PoseTrajectory3D):
seconds_from_start = [t - traj_est.timestamps[0]
for t in traj_est.timestamps]
ape_result.add_np_array("seconds_from_start", seconds_from_start)
ape_result.add_np_array("timestamps", traj_est.timestamps)
return ape_result
def compare_using_APE(ref_file, est_file, use_aligned_trajectories=False):
"""
Compare two files using EVO API. Using the APE metric.
:param ref_file:
:param est_file:
:param use_aligned_trajectories: True to align before comparing. False to leave original data as it is.
:return:
"""
# Load trajectories
traj_ref = file_interface.read_tum_trajectory_file(ref_file)
traj_est = file_interface.read_tum_trajectory_file(est_file)
# Sinchronize trajectories by timestamps
max_diff = 0.01
traj_ref, traj_est = sync.associate_trajectories(traj_ref, traj_est, max_diff)
# -------------EVO_APE-------------
# Settings
pose_relation = metrics.PoseRelation.translation_part
use_aligned_trajectories = False # OPTION -va on the scripts. Is related to Uleyamas alignment
# The aligned trajectories can be used if we want it to
if use_aligned_trajectories:
# Align trajectories with Uleyamas algorithm
try:
traj_est_aligned = trajectory.align_trajectory(traj_est, traj_ref, correct_scale=False,
correct_only_scale=False)
data = (traj_ref, traj_est_aligned)
except GeometryException:
print("Couldnt align with Uleyamas algorithm...")
data = (traj_ref, traj_est)
else:
data = (traj_ref, traj_est)
ape_metric = metrics.APE(pose_relation) # APE with only pose is in reality ATE (Absolute trajectory error) instead of APE (abs. pose error)
ape_metric.process_data(data)
# Get all stadistics in a dict
ape_stats = ape_metric.get_all_statistics()
return ape_stats
def ape(traj_ref,
traj_est,
pose_relation,
align=False,
correct_scale=False,
align_origin=False,
ref_name="reference",
est_name="estimate"):
''' Based on main_ape.py
'''
traj_ref, traj_est = sync.associate_trajectories(traj_ref, traj_est)
# Align the trajectories.
only_scale = correct_scale and not align
if align or correct_scale:
traj_est = trajectory.align_trajectory(traj_est, traj_ref,
correct_scale, only_scale)
elif align_origin:
traj_est = trajectory.align_trajectory_origin(traj_est, traj_ref)
# Calculate APE.
data = (traj_ref, traj_est)
ape_metric = APE(pose_relation)
ape_metric.process_data(data)
ape_result = ape_metric.get_result(ref_name, est_name)
ape_result.add_trajectory(ref_name, traj_ref)
ape_result.add_trajectory(est_name, traj_est)
if isinstance(traj_est, trajectory.PoseTrajectory3D):
seconds_from_start = [
t - traj_est.timestamps[0] for t in traj_est.timestamps
]
ape_result.add_np_array("seconds_from_start", seconds_from_start)
ape_result.add_np_array("timestamps", traj_est.timestamps)
return ape_result
#!/usr/bin/env python
from __future__ import print_function
print("loading required evo modules")
from evo.core import trajectory, sync, metrics
from evo.tools import file_interface
print("loading trajectories")
traj_ref = file_interface.read_tum_trajectory_file("../../test/data/fr2_desk_groundtruth.txt")
traj_est = file_interface.read_tum_trajectory_file("../../test/data/fr2_desk_ORB.txt")
print("registering and aligning trajectories")
traj_ref, traj_est = sync.associate_trajectories(traj_ref, traj_est)
traj_est = trajectory.align_trajectory(traj_est, traj_ref, correct_scale=False)
print("calculating APE")
data = (traj_ref, traj_est)
ape_metric = metrics.APE(metrics.PoseRelation.translation_part)
ape_metric.process_data(data)
ape_statistics = ape_metric.get_all_statistics()
print("mean:", ape_statistics["mean"])
print("loading plot modules")
from evo.tools import plot
import matplotlib.pyplot as plt
print("plotting")
plot_collection = plot.PlotCollection("Example")
# metric values
fig_1 = plt.figure(figsize=(8, 8))
import numpy as np
import matplotlib.pyplot as plt
logger = logging.getLogger("evo")
log.configure_logging(verbose=True)
traj_ref = file_interface.read_kitti_poses_file("../test/data/KITTI_00_gt.txt")
traj_est = file_interface.read_kitti_poses_file("../test/data/KITTI_00_ORB.txt")
# add artificial Sim(3) transformation
traj_est.transform(lie.se3(np.eye(3), [0, 0, 0]))
traj_est.scale(0.5)
logger.info("\nUmeyama alignment without scaling")
traj_est_aligned = trajectory.align_trajectory(traj_est, traj_ref)
logger.info("\nUmeyama alignment with scaling")
traj_est_aligned_scaled = trajectory.align_trajectory(
traj_est, traj_ref, correct_scale=True)
logger.info("\nUmeyama alignment with scaling only")
traj_est_aligned_only_scaled = trajectory.align_trajectory(
traj_est, traj_ref, correct_only_scale=True)
fig = plt.figure(figsize=(8, 8))
plot_mode = plot.PlotMode.xz
ax = plot.prepare_axis(fig, plot_mode, subplot_arg='221')
plot.traj(ax, plot_mode, traj_ref, '--', 'gray')
plot.traj(ax, plot_mode, traj_est, '-', 'blue')
def run_analysis(traj_ref_path,
traj_est_path,
segments,
save_results,
display_plot,
save_plots,
save_folder,
confirm_overwrite=False,
dataset_name="",
discard_n_start_poses=0,
discard_n_end_poses=0):
""" Run analysis on given trajectories, saves plots on given path:
:param traj_ref_path: path to the reference (ground truth) trajectory.
:param traj_est_path: path to the estimated trajectory.
:param save_results: saves APE, and RPE per segment results.
:param save_plots: whether to save the plots.
:param save_folder: where to save the plots.
:param confirm_overwrite: whether to confirm overwriting plots or not.
:param dataset_name: optional param, to allow setting the same scale on different plots.
"""
# Load trajectories.
from evo.tools import file_interface
traj_ref = None
try:
traj_ref = file_interface.read_euroc_csv_trajectory(
traj_ref_path) # TODO make it non-euroc specific.
except file_interface.FileInterfaceException as e:
raise Exception(
"\033[91mMissing ground truth csv! \033[93m {}.".format(e))
traj_est = None
try:
traj_est = file_interface.read_swe_csv_trajectory(traj_est_path)
except file_interface.FileInterfaceException as e:
log.info(e)
raise Exception("\033[91mMissing vio output csv.\033[99m")
evt.print_purple("Registering trajectories")
traj_ref, traj_est = sync.associate_trajectories(traj_ref, traj_est)
evt.print_purple("Aligning trajectories")
traj_est = trajectory.align_trajectory(
traj_est,
traj_ref,
correct_scale=False,
discard_n_start_poses=int(discard_n_start_poses),
discard_n_end_poses=int(discard_n_end_poses))
num_of_poses = traj_est.num_poses
traj_est.reduce_to_ids(
range(int(discard_n_start_poses),
int(num_of_poses - discard_n_end_poses), 1))
traj_ref.reduce_to_ids(
range(int(discard_n_start_poses),
int(num_of_poses - discard_n_end_poses), 1))
results = dict()
evt.print_purple("Calculating APE translation part")
data = (traj_ref, traj_est)
ape_metric = metrics.APE(metrics.PoseRelation.translation_part)
ape_metric.process_data(data)
ape_result = ape_metric.get_result()
results["absolute_errors"] = ape_result
log.info(ape_result.pretty_str(info=True))
# TODO(Toni): Save RPE computation results rather than the statistics
# you can compute statistics later...
evt.print_purple("Calculating RPE translation part for plotting")
rpe_metric_trans = metrics.RPE(metrics.PoseRelation.translation_part, 1.0,
metrics.Unit.frames, 0.0, False)
rpe_metric_trans.process_data(data)
rpe_stats_trans = rpe_metric_trans.get_all_statistics()
log.info("mean: %f" % rpe_stats_trans["mean"])
evt.print_purple("Calculating RPE rotation angle for plotting")
rpe_metric_rot = metrics.RPE(metrics.PoseRelation.rotation_angle_deg, 1.0,
metrics.Unit.frames, 1.0, False)
rpe_metric_rot.process_data(data)
rpe_stats_rot = rpe_metric_rot.get_all_statistics()
log.info("mean: %f" % rpe_stats_rot["mean"])
results["relative_errors"] = dict()
# Read segments file
for segment in segments:
results["relative_errors"][segment] = dict()
evt.print_purple("RPE analysis of segment: %d" % segment)
evt.print_lightpurple("Calculating RPE segment translation part")
rpe_segment_metric_trans = metrics.RPE(
metrics.PoseRelation.translation_part, float(segment),
metrics.Unit.meters, 0.01, True)
rpe_segment_metric_trans.process_data(data)
rpe_segment_stats_trans = rpe_segment_metric_trans.get_all_statistics()
results["relative_errors"][segment][
"rpe_trans"] = rpe_segment_stats_trans
# print(rpe_segment_stats_trans)
# print("mean:", rpe_segment_stats_trans["mean"])
evt.print_lightpurple("Calculating RPE segment rotation angle")
rpe_segment_metric_rot = metrics.RPE(
metrics.PoseRelation.rotation_angle_deg, float(segment),
metrics.Unit.meters, 0.01, True)
rpe_segment_metric_rot.process_data(data)
rpe_segment_stats_rot = rpe_segment_metric_rot.get_all_statistics()
results["relative_errors"][segment]["rpe_rot"] = rpe_segment_stats_rot
# print(rpe_segment_stats_rot)
# print("mean:", rpe_segment_stats_rot["mean"])
if save_results:
# Save results file
results_file = os.path.join(save_folder, 'results.yaml')
evt.print_green("Saving analysis results to: %s" % results_file)
with open(results_file, 'w') as outfile:
if confirm_overwrite:
if evt.user.check_and_confirm_overwrite(results_file):
outfile.write(yaml.dump(results, default_flow_style=False))
else:
log.info("Not overwritting results.")
else:
outfile.write(yaml.dump(results, default_flow_style=False))
# For each segment in segments file
# Calculate rpe with delta = segment in meters with all-pairs set to True
# Calculate max, min, rmse, mean, median etc
# Plot boxplot, or those cumulative figures you see in evo (like demographic plots)
if display_plot or save_plots:
evt.print_green("Plotting:")
log.info(dataset_name)
plot_collection = plot.PlotCollection("Example")
# metric values
fig_1 = plt.figure(figsize=(8, 8))
ymax = -1
if dataset_name is not "" and FIX_MAX_Y:
ymax = Y_MAX_APE_TRANS[dataset_name]
ape_statistics = ape_metric.get_all_statistics()
plot.error_array(
fig_1,
ape_metric.error,
statistics=ape_statistics,
name="APE translation",
title="" #str(ape_metric)
,
xlabel="Keyframe index [-]",
ylabel="APE translation [m]",
y_min=0.0,
y_max=ymax)
plot_collection.add_figure("APE_translation", fig_1)
# trajectory colormapped with error
fig_2 = plt.figure(figsize=(8, 8))
plot_mode = plot.PlotMode.xy
ax = plot.prepare_axis(fig_2, plot_mode)
plot.traj(ax, plot_mode, traj_ref, '--', 'gray', 'reference')
plot.traj_colormap(ax,
traj_est,
ape_metric.error,
plot_mode,
min_map=0.0,
max_map=math.ceil(ape_statistics['max'] * 10) / 10,
title="ATE mapped onto trajectory [m]")
plot_collection.add_figure("APE_translation_trajectory_error", fig_2)
# RPE
## Trans
### metric values
fig_3 = plt.figure(figsize=(8, 8))
if dataset_name is not "" and FIX_MAX_Y:
ymax = Y_MAX_RPE_TRANS[dataset_name]
plot.error_array(
fig_3,
rpe_metric_trans.error,
statistics=rpe_stats_trans,
name="RPE translation",
title="" #str(rpe_metric_trans)
,
xlabel="Keyframe index [-]",
ylabel="RPE translation [m]",
y_max=ymax)
plot_collection.add_figure("RPE_translation", fig_3)
### trajectory colormapped with error
fig_4 = plt.figure(figsize=(8, 8))
plot_mode = plot.PlotMode.xy
ax = plot.prepare_axis(fig_4, plot_mode)
traj_ref_trans = copy.deepcopy(traj_ref)
traj_ref_trans.reduce_to_ids(rpe_metric_trans.delta_ids)
traj_est_trans = copy.deepcopy(traj_est)
traj_est_trans.reduce_to_ids(rpe_metric_trans.delta_ids)
plot.traj(ax, plot_mode, traj_ref_trans, '--', 'gray', 'Reference')
plot.traj_colormap(
ax,
traj_est_trans,
rpe_metric_trans.error,
plot_mode,
min_map=0.0,
max_map=math.ceil(rpe_stats_trans['max'] * 10) / 10,
title="RPE translation error mapped onto trajectory [m]")
plot_collection.add_figure("RPE_translation_trajectory_error", fig_4)
## Rot
### metric values
fig_5 = plt.figure(figsize=(8, 8))
if dataset_name is not "" and FIX_MAX_Y:
ymax = Y_MAX_RPE_ROT[dataset_name]
plot.error_array(
fig_5,
rpe_metric_rot.error,
statistics=rpe_stats_rot,
name="RPE rotation error",
title="" #str(rpe_metric_rot)
,
xlabel="Keyframe index [-]",
ylabel="RPE rotation [deg]",
y_max=ymax)
plot_collection.add_figure("RPE_rotation", fig_5)
### trajectory colormapped with error
fig_6 = plt.figure(figsize=(8, 8))
plot_mode = plot.PlotMode.xy
ax = plot.prepare_axis(fig_6, plot_mode)
traj_ref_rot = copy.deepcopy(traj_ref)
traj_ref_rot.reduce_to_ids(rpe_metric_rot.delta_ids)
traj_est_rot = copy.deepcopy(traj_est)
traj_est_rot.reduce_to_ids(rpe_metric_rot.delta_ids)
plot.traj(ax, plot_mode, traj_ref_rot, '--', 'gray', 'Reference')
plot.traj_colormap(
ax,
traj_est_rot,
rpe_metric_rot.error,
plot_mode,
min_map=0.0,
max_map=math.ceil(rpe_stats_rot['max'] * 10) / 10,
title="RPE rotation error mapped onto trajectory [deg]")
plot_collection.add_figure("RPE_rotation_trajectory_error", fig_6)
if display_plot:
evt.print_green("Displaying plots.")
plot_collection.show()
if save_plots:
evt.print_green("Saving plots to: ")
log.info(save_folder)
# Config output format (pdf, eps, ...) using evo_config...
plot_collection.export(os.path.join(save_folder, "plots.eps"),
False)
plot_collection.export(os.path.join(save_folder, "plots.pdf"),
False)
# load trajectories (examples from TUM benchmark)
ref_file = "data/freiburg1_xyz-groundtruth.txt"
est_file = "data/freiburg1_xyz-rgbdslam_drift.txt"
max_diff = 0.01
offset = 0.0
start = time.clock()
traj_ref, traj_est = file_interface.load_assoc_tum_trajectories(
est_file,
ref_file,
max_diff,
offset,
)
# align trajectories
print("\naligning trajectories...")
traj_est = trajectory.align_trajectory(traj_est, traj_ref)
stop = time.clock()
load_time = stop - start
print("elapsed time for trajectory loading (seconds): \t", "{0:.6f}".format(load_time))
"""
test absolute pose error algorithms
for different types of pose relations
"""
for pose_relation in metrics.PoseRelation:
print("\n------------------------------------------------------------------\n")
data = (traj_ref, traj_est)
start = time.clock()
ape_metric = metrics.APE(pose_relation)
def ape(traj_ref_list,
traj_est_list,
pose_relation,
align=False,
correct_scale=False,
align_origin=False,
ref_name=["reference"],
est_name=["estimate"]):
from evo.core import metrics
from evo.core import trajectory
# Align the trajectories.
only_scale = correct_scale and not align
if align or correct_scale:
logger.debug(SEP)
for i in range(len(traj_ref_list)):
print(traj_est_list[i], traj_ref_list[i])
traj_est_list[i] = trajectory.align_trajectory(
traj_est_list[i], traj_ref_list[i], correct_scale, only_scale)
elif align_origin:
logger.debug(SEP)
for i in range(len(traj_ref_list)):
traj_est_list[i] = trajectory.align_trajectory_origin(
traj_est_list[i], traj_ref_list[i])
# Calculate APE.
logger.debug(SEP)
data = [(traj_ref_list[i], traj_est_list[i])
for i in range(len(traj_ref_list))]
ape_metric = [
metrics.APE(pose_relation) for i in range(len(traj_ref_list))
]
for i in range(len(traj_ref_list)):
ape_metric[i].process_data(data[i])
ape_result = [
ape_metric[i].get_result(ref_name[i], est_name[i])
for i in range(len(traj_ref_list))
]
for i in range(len(traj_ref_list)):
title = str(ape_metric[i])
if align and not correct_scale:
title += "\n(with SE(3) Umeyama alignment)"
elif align and correct_scale:
title += "\n(with Sim(3) Umeyama alignment)"
elif only_scale:
title += "\n(scale corrected)"
elif align_origin:
title += "\n(with origin alignment)"
else:
title += "\n(not aligned)"
ape_result[i].info["title"] = title
logger.debug(SEP)
logger.info(ape_result[i].pretty_str())
ape_result[i].add_trajectory(ref_name[i], traj_ref_list[i])
ape_result[i].add_trajectory(est_name[i], traj_est_list[i])
if isinstance(traj_est_list[i], trajectory.PoseTrajectory3D):
seconds_from_start = [
t - traj_est_list[i].timestamps[0]
for t in traj_est_list[i].timestamps
]
ape_result[i].add_np_array("seconds_from_start",
seconds_from_start)
ape_result[i].add_np_array("timestamps",
traj_est_list[i].timestamps)
return ape_result
def get_and_save_results_from_folder(folder_with_predicted_poses,category):
global args
global kitti_eval_tool
global folder_with_gt_poses
global output_folder
global t
global results
values_for_excel = []
columns_for_excel = []
type_of_statistics = 'mean'
for filename in sorted(os.listdir(folder_with_predicted_poses)):
if not(os.path.exists(os.path.join(folder_with_gt_poses, filename))):
print("file with gt poses doesn't exist for "+filename)
continue
if filename.find('.txt') == -1:
continue
seq_results = {}
seq_results['name_seq'] = filename[:filename.rfind('.')]
seq_results['category'] = category
folder_name = seq_results['category']
seq_results['metrics'] = {}
seq_results['lost'] = False
os.makedirs(os.path.join(output_folder, folder_name), exist_ok=True)
output_folder_seq = os.path.join(output_folder, folder_name, filename[:filename.rfind('.')])
os.makedirs(output_folder_seq, exist_ok=True)
if os.path.isfile(os.path.join(output_folder, folder_name,"results.txt")):
file_results_txt = open(os.path.join(output_folder, folder_name,"results.txt"), "a")
else:
file_results_txt = open(os.path.join(output_folder, folder_name,"results.txt"), "w")
file_results_txt.write("translation_error(%) rotation_error(deg/m) ATE(m) APE_translation_error_median(m) APE_rotation_error_median(deg) dst_to_trgt\n")
# -------------------------------------Getting results---------------------------------------------------
if args.gt_format == 'kitti':
traj_ref = file_interface.read_kitti_poses_file(os.path.join(folder_with_gt_poses, filename))
if args.gt_format == 'tum':
traj_ref = file_interface.read_tum_trajectory_file(os.path.join(folder_with_gt_poses, filename))
seq_results["length_of_ref_traj"] = traj_ref.path_length
end_time_gt = traj_ref.get_infos()["t_end (s)"]
if args.gt_format == 'euroc':
traj_ref = file_interface.read_euroc_csv_trajectory(os.path.join(folder_with_gt_poses, filename))
if args.result_format == 'kitti':
traj_est = file_interface.read_kitti_poses_file(os.path.join(folder_with_predicted_poses, filename))
if args.result_format == 'tum':
traj_est = file_interface.read_tum_trajectory_file(os.path.join(folder_with_predicted_poses, filename))
seq_results["length_of_estimated_traj"] = traj_est.path_length
if args.result_format == 'euroc':
traj_est = file_interface.read_euroc_csv_trajectory(os.path.join(folder_with_predicted_poses, filename))
if args.result_format == 'tum' and args.gt_format == 'tum':
seq_results["num_gt_poses"] = traj_ref.num_poses
seq_results["num_predicted_poses"] = traj_est.num_poses
traj_ref, traj_est = sync.associate_trajectories(traj_ref, traj_est, args.max_diff)
end_time_est = traj_est.get_infos()["t_end (s)"]
if (abs(end_time_est - end_time_gt) > 0.2) or (traj_est.get_infos()["t_start (s)"] > 0.2):
print('LOST in track '+filename[:filename.rfind('.')])
seq_results['lost'] = True
results.append(seq_results)
t.update(1)
continue
if args.alignment != None:
traj_est = trajectory.align_trajectory(traj_est, traj_ref, correct_scale=args.alignment.find("scale") != -1, correct_only_scale=args.alignment=="scale")
trajectory.align_trajectory_origin(traj_est, traj_ref)
data = (traj_ref, traj_est)
ape_metric_translation = metrics.APE(metrics.PoseRelation.translation_part)
ape_metric_rotation = metrics.APE(metrics.PoseRelation.rotation_angle_deg)
ape_metric_translation.process_data(data)
ape_metric_rotation.process_data(data)
ape_translation_statistics = ape_metric_translation.get_all_statistics()
ape_rotation_statistics = ape_metric_rotation.get_all_statistics()
ape_translation_statistics_plot = copy.deepcopy(ape_translation_statistics)
ape_rotation_statistics_plot = copy.deepcopy(ape_rotation_statistics)
ape_translation_statistics_plot.pop('sse')
ape_translation_statistics_plot.pop('std')
ape_translation_statistics_plot.pop('min')
ape_translation_statistics_plot.pop('max')
ape_rotation_statistics_plot.pop('sse')
ape_rotation_statistics_plot.pop('std')
ape_rotation_statistics_plot.pop('min')
ape_rotation_statistics_plot.pop('max')
kitti_trans_err, kitti_rot_err, ate = kitti_eval_tool.eval(traj_ref.poses_se3,
traj_est.poses_se3,
alignment=None)
#---------------------------------adding results to variable seq_results for excel -----------------------------
seq_results['metrics']['dist_to_trgt'] = traj_est.get_infos()['pos_end (m)'] - traj_ref.get_infos()['pos_end (m)']
seq_results['metrics']['dist_to_trgt'] = np.sum(np.array(seq_results['metrics']['dist_to_trgt'])**2)**0.5
seq_results['metrics']["Kitti trans err (%)"] = kitti_trans_err
seq_results['metrics']["Kitti rot err (deg/m)"] = kitti_rot_err
seq_results['metrics']["ATE (m)"] = ate
seq_results['metrics']["APE(trans err) median (m)"] = ape_translation_statistics["median"]
seq_results['metrics']["APE(rot err) median (deg)"] = ape_rotation_statistics["median"]
#--------------------------------------------------------------------------------------------------------
#-------------------------------------------------------------------------------------------------------
# --------------------------------printing results into console----------------------------------------------
print('Results for "'+filename+'":')
print('Kitti average translational error (%): {:.7f}'.format(kitti_trans_err))
print('Kitti average rotational error (deg/m): {:.7f}'.format(kitti_rot_err))
print('ATE (m): {:.7f}'.format(ate))
print('APE(translation error) median (m): {:.7f}'.format(ape_translation_statistics["median"]))
print('APE(rotation error) median (deg): {:.7f}'.format(ape_rotation_statistics["median"]))
print('distance to target on the last frame: {:.7f}'.format(seq_results['metrics']['dist_to_trgt']))
#------------------------------------------------------------------------------------------------------------
#---------------------------------Saving results into overall results text file------------------------------
file_results_txt.write('{:<24} '.format(filename[:filename.rfind('.')]))
file_results_txt.write('{:>7.4f} '.format(kitti_trans_err))
file_results_txt.write('{:>7.4f} '.format(kitti_rot_err))
file_results_txt.write('{:>7.4f} '.format(ate))
file_results_txt.write('{:>7.4f} '.format(ape_translation_statistics["median"]))
file_results_txt.write('{:>7.4f} '.format(ape_rotation_statistics["median"]))
file_results_txt.write('{:>7.4f}\n'.format(seq_results['metrics']['dist_to_trgt']))
#------------------------------------------------------------------------------------------------------------
# --------------------------------Saving metrics to text file for one track----------------------------------
txt_filename = filename[:filename.rfind('.')]+"_metrics.txt"
with open(os.path.join(output_folder_seq, txt_filename), "w") as txt_file:
txt_file.write('Kitti average translational error (%): {:.7f}\n'.format(kitti_trans_err))
txt_file.write('Kitti average rotational error (deg/m): {:.7f}\n'.format(kitti_rot_err))
txt_file.write('ATE (m): {:.7f}\n'.format(ate))
txt_file.write('APE(translation error) median (m): {:.7f}\n'.format(ape_translation_statistics["median"]))
txt_file.write('APE(rotation error) median (deg): {:.7f}\n'.format(ape_rotation_statistics["median"]))
txt_file.write('Distance to target on the last frame: {:.7f}\n'.format(seq_results['metrics']['dist_to_trgt']))
#---------------------------------------------------------------------------------------------------------
# ---------------------------------Saving values of errors for each frame to text file------------------------
# ------------------------------------------for translation errors----------------------------------------
txt_filename = filename[:filename.rfind('.')]+"_APE(translation)_errors.txt"
output_folder_seq_translation = os.path.join(output_folder_seq,"translation")
output_folder_seq_rotation = os.path.join(output_folder_seq,"rotation")
os.makedirs(output_folder_seq_translation, exist_ok=True)
os.makedirs(output_folder_seq_rotation, exist_ok=True)
with open(os.path.join(output_folder_seq_translation, txt_filename), "w") as txt_file:
for error in ape_metric_translation.error:
txt_file.write('{:.10f}\n'.format(error))
# -----------------------------------------for rotation degree errors--------------------------------------
txt_filename = filename[:filename.rfind('.')]+"_APE(rotation_deg)_errors.txt"
with open(os.path.join(output_folder_seq_rotation, txt_filename), "w") as txt_file:
for error in ape_metric_rotation.error:
txt_file.write('{:.10f}\n'.format(error))
#----------------------------------------------------------------------------------------------------------
# ---------------------------------------Saving plot of errors of each frame------------------------------
# ------------------------------------------for translation errors----------------------------------------
plot_collection = plot.PlotCollection("Example")
fig_1 = plt.figure(figsize=(8, 8))
plot.error_array(fig_1, ape_metric_translation.error,
name="APE", title=str(ape_metric_translation), xlabel="Index of frame", ylabel='Error')
plot_collection.add_figure("raw", fig_1)
plot_filename = filename[:filename.rfind('.')]+"_APE(translation)_errors.png"
plt.savefig(os.path.join(output_folder_seq_translation, plot_filename))
plt.close(fig_1)
# -----------------------------------------for rotation degree errors--------------------------------------
plot_collection = plot.PlotCollection("Example")
fig_1 = plt.figure(figsize=(8, 8))
plot.error_array(fig_1, ape_metric_rotation.error,
name="APE", title=str(ape_metric_rotation), xlabel="Index of frame", ylabel='Error')
plot_collection.add_figure("raw", fig_1)
plot_filename = filename[:filename.rfind('.')]+"_APE(rotation)_errors.png"
plt.savefig(os.path.join(output_folder_seq_rotation,plot_filename))
plt.close(fig_1)
#-----------------------------------------------------------------------------------------------------------
# -----------------------------------------Saving trajectory plot-------------------------------------------
# ------------------------------------------for translation errors----------------------------------------
fig_2 = plt.figure(figsize=(8, 8))
ax = plot.prepare_axis(fig_2, plot_mode)
plot.traj(ax, plot_mode, traj_ref, '--', 'gray', 'reference')
plot.traj_colormap( ax, traj_est, ape_metric_translation.error, plot_mode,
min_map=ape_translation_statistics["min"],
max_map=ape_translation_statistics["max"], title="APE translation mapped onto trajectory")
plot_collection.add_figure("traj (error)", fig_2)
plot_filename = filename[:filename.rfind('.')]+"_APE(translation)_map.png"
plt.savefig(os.path.join(output_folder_seq_translation,plot_filename))
plt.close(fig_2)
# -----------------------------------------for rotation degree errors--------------------------------------
fig_2 = plt.figure(figsize=(8, 8))
ax = plot.prepare_axis(fig_2, plot_mode)
plot.traj(ax, plot_mode, traj_ref, '--', 'gray', 'reference')
plot.traj_colormap( ax, traj_est, ape_metric_rotation.error, plot_mode,
min_map=ape_rotation_statistics["min"],
max_map=ape_rotation_statistics["max"], title="APE rotation mapped onto trajectory")
plot_collection.add_figure("traj (error)", fig_2)
plot_filename = filename[:filename.rfind('.')]+"_APE(rotation)_map.png"
plt.savefig(os.path.join(output_folder_seq_rotation,plot_filename))
plt.close(fig_2)
#-----------------------------------------------------------------------------------------------------------
print()
active_worksheet = wb['sheet1']
thin = Side(border_style="thin", color="000000")
thick = Side(border_style="thick", color="000000")
medium = Side(border_style="medium", color="000000")
font_header = Font(name='Arial',
size=10,
bold=True,
italic=False,
vertAlign=None,
underline='none',
strike=False,
color='FF000000')
font_values = Font(name='Arial',
size=10,
bold=False,
italic=False,
vertAlign=None,
underline='none',
strike=False,
color='FF000000')
active_worksheet.row_dimensions[2].height = 35
file_results_txt.close()
results.append(seq_results)
t.update(1)
def main_ape(traj_ref,
traj_est,
pose_relation,
align=True,
correct_scale=False,
ref_name="",
est_name="",
show_plot=False,
save_plot=None,
plot_mode=None,
save_results=None,
no_warnings=False,
serialize_plot=None,
plot_colormap_max=None,
plot_colormap_min=None,
plot_colormap_max_percentile=None):
from evo.core import metrics, result
from evo.core import trajectory
from evo.tools import file_interface
from evo.tools.settings import SETTINGS
import numpy as np
only_scale = correct_scale and not align
if align or correct_scale:
logger.debug(SEP)
if only_scale:
logger.debug("correcting scale...")
else:
logger.debug("aligning using Umeyama's method..." +
(" (with scale correction)" if correct_scale else ""))
traj_est = trajectory.align_trajectory(traj_est, traj_ref,
correct_scale, only_scale)
logger.debug(SEP)
# calculate APE
data = (traj_ref, traj_est)
ape_metric = metrics.APE(pose_relation)
ape_metric.process_data(data)
ape_statistics = ape_metric.get_all_statistics()
title = str(ape_metric)
if align and not correct_scale:
title += "\n(with SE(3) Umeyama alignment)"
elif align and correct_scale:
title += "\n(with Sim(3) Umeyama alignment)"
elif only_scale:
title += "\n(scale corrected)"
else:
title += "\n(not aligned)"
ape_result = ape_metric.get_result(ref_name, est_name)
logger.debug(SEP)
logger.info(ape_result.pretty_str())
if isinstance(traj_est, trajectory.PoseTrajectory3D):
seconds_from_start = [
t - traj_est.timestamps[0] for t in traj_est.timestamps
]
ape_result.add_np_array("seconds_from_start", seconds_from_start)
ape_result.add_np_array("timestamps", traj_est.timestamps)
else:
seconds_from_start = None
if show_plot or save_plot or save_results or serialize_plot:
if show_plot or save_plot or serialize_plot:
from evo.tools import plot
import matplotlib.pyplot as plt
logger.debug(SEP)
logger.debug("plotting results... ")
fig1 = plt.figure(figsize=(SETTINGS.plot_figsize[0],
SETTINGS.plot_figsize[1]))
# metric values
plot.error_array(
fig1,
ape_metric.error,
x_array=seconds_from_start,
statistics=ape_statistics,
name="APE" + (" (" + ape_metric.unit.value + ")")
if ape_metric.unit else "",
title=title,
xlabel="$t$ (s)" if seconds_from_start else "index")
# info text
if SETTINGS.plot_info_text and est_name and ref_name:
ax = fig1.gca()
ax.text(0,
-0.12,
"estimate: " + est_name + "\nreference: " + ref_name,
transform=ax.transAxes,
fontsize=8,
color="gray")
# trajectory colormapped
fig2 = plt.figure(figsize=(SETTINGS.plot_figsize[0],
SETTINGS.plot_figsize[1]))
plot_mode = plot_mode if plot_mode is not None else plot.PlotMode.xyz
ax = plot.prepare_axis(fig2, plot_mode)
plot.traj(ax,
plot_mode,
traj_ref,
'--',
'gray',
'reference',
alpha=0.0 if SETTINGS.plot_hideref else 1.0)
min_map = ape_statistics[
"min"] if plot_colormap_min is None else plot_colormap_min
if plot_colormap_max_percentile is not None:
max_map = np.percentile(ape_result.np_arrays["error_array"],
plot_colormap_max_percentile)
else:
max_map = ape_statistics[
"max"] if plot_colormap_max is None else plot_colormap_max
plot.traj_colormap(ax,
traj_est,
ape_metric.error,
plot_mode,
min_map=min_map,
max_map=max_map,
title="APE mapped onto trajectory")
fig2.axes.append(ax)
plot_collection = plot.PlotCollection(title)
plot_collection.add_figure("raw", fig1)
plot_collection.add_figure("map", fig2)
if show_plot:
plot_collection.show()
if save_plot:
plot_collection.export(save_plot,
confirm_overwrite=not no_warnings)
if serialize_plot:
logger.debug(SEP)
plot_collection.serialize(serialize_plot,
confirm_overwrite=not no_warnings)
if save_results:
logger.debug(SEP)
if SETTINGS.save_traj_in_zip:
ape_result.add_trajectory("traj_ref", traj_ref)
ape_result.add_trajectory("traj_est", traj_est)
file_interface.save_res_file(save_results,
ape_result,
confirm_overwrite=not no_warnings)
return ape_result
def run_analysis(self,
traj_ref_path,
traj_vio_path,
traj_pgo_path,
segments,
dataset_name="",
discard_n_start_poses=0,
discard_n_end_poses=0):
""" Analyze data from a set of trajectory csv files.
Args:
traj_ref_path: string representing filepath of the reference (ground-truth) trajectory.
traj_vio_path: string representing filepath of the vio estimated trajectory.
traj_pgo_path: string representing filepath of the pgo estimated trajectory.
segments: list of segments for RPE calculation, defined in the experiments yaml file.
dataset_name: string representing the dataset's name
discard_n_start_poses: int representing number of poses to discard from start of analysis.
discard_n_end_poses: int representing the number of poses to discard from end of analysis.
"""
import copy
# Mind that traj_est_pgo might be None
traj_ref, traj_est_vio, traj_est_pgo = self.read_traj_files(
traj_ref_path, traj_vio_path, traj_pgo_path)
# We copy to distinguish from the pgo version that may be created
traj_ref_vio = copy.deepcopy(traj_ref)
# Register and align trajectories:
evt.print_purple("Registering and aligning trajectories")
traj_ref_vio, traj_est_vio = sync.associate_trajectories(
traj_ref_vio, traj_est_vio)
traj_est_vio = trajectory.align_trajectory(
traj_est_vio,
traj_ref_vio,
correct_scale=False,
discard_n_start_poses=int(discard_n_start_poses),
discard_n_end_poses=int(discard_n_end_poses))
# We do the same for the PGO trajectory if needed:
traj_ref_pgo = None
if traj_est_pgo is not None:
traj_ref_pgo = copy.deepcopy(traj_ref)
traj_ref_pgo, traj_est_pgo = sync.associate_trajectories(
traj_ref_pgo, traj_est_pgo)
traj_est_pgo = trajectory.align_trajectory(
traj_est_pgo,
traj_ref_pgo,
correct_scale=False,
discard_n_start_poses=int(discard_n_start_poses),
discard_n_end_poses=int(discard_n_end_poses))
# We need to pick the lowest num_poses before doing any computation:
num_of_poses = traj_est_vio.num_poses
if traj_est_pgo is not None:
num_of_poses = min(num_of_poses, traj_est_pgo.num_poses)
traj_est_pgo.reduce_to_ids(
range(int(discard_n_start_poses),
int(num_of_poses - discard_n_end_poses), 1))
traj_ref_pgo.reduce_to_ids(
range(int(discard_n_start_poses),
int(num_of_poses - discard_n_end_poses), 1))
traj_est_vio.reduce_to_ids(
range(int(discard_n_start_poses),
int(num_of_poses - discard_n_end_poses), 1))
traj_ref_vio.reduce_to_ids(
range(int(discard_n_start_poses),
int(num_of_poses - discard_n_end_poses), 1))
# Calculate all metrics:
(ape_metric_vio, rpe_metric_trans_vio, rpe_metric_rot_vio,
results_vio) = self.process_trajectory_data(traj_ref_vio,
traj_est_vio, segments,
True)
# We do the same for the pgo trajectory if needed:
ape_metric_pgo = None
rpe_metric_trans_pgo = None
rpe_metric_rot_pgo = None
results_pgo = None
if traj_est_pgo is not None:
(ape_metric_pgo, rpe_metric_trans_pgo, rpe_metric_rot_pgo,
results_pgo) = self.process_trajectory_data(
traj_ref_pgo, traj_est_pgo, segments, False)
# Generate plots for return:
plot_collection = None
if self.display_plots or self.save_plots:
evt.print_green("Plotting:")
log.info(dataset_name)
plot_collection = plot.PlotCollection("Example")
if traj_est_pgo is not None:
# APE Metric Plot:
plot_collection.add_figure(
"PGO_APE_translation",
plot_metric(ape_metric_pgo, "PGO + VIO APE Translation"))
# Trajectory Colormapped with ATE Plot:
plot_collection.add_figure(
"PGO_APE_translation_trajectory_error",
plot_traj_colormap_ape(
ape_metric_pgo, traj_ref_pgo, traj_est_vio,
traj_est_pgo, "PGO + VIO ATE Mapped Onto Trajectory"))
# RPE Translation Metric Plot:
plot_collection.add_figure(
"PGO_RPE_translation",
plot_metric(rpe_metric_trans_pgo,
"PGO + VIO RPE Translation"))
# Trajectory Colormapped with RTE Plot:
plot_collection.add_figure(
"PGO_RPE_translation_trajectory_error",
plot_traj_colormap_rpe(
rpe_metric_trans_pgo, traj_ref_pgo, traj_est_vio,
traj_est_pgo,
"PGO + VIO RPE Translation Error Mapped Onto Trajectory"
))
# RPE Rotation Metric Plot:
plot_collection.add_figure(
"PGO_RPE_Rotation",
plot_metric(rpe_metric_rot_pgo, "PGO + VIO RPE Rotation"))
# Trajectory Colormapped with RTE Plot:
plot_collection.add_figure(
"PGO_RPE_rotation_trajectory_error",
plot_traj_colormap_rpe(
rpe_metric_rot_pgo, traj_ref_pgo, traj_est_vio,
traj_est_pgo,
"PGO + VIO RPE Rotation Error Mapped Onto Trajectory"))
# Plot VIO results
plot_collection.add_figure(
"VIO_APE_translation",
plot_metric(ape_metric_vio, "VIO APE Translation"))
plot_collection.add_figure(
"VIO_APE_translation_trajectory_error",
plot_traj_colormap_ape(ape_metric_vio, traj_ref_vio,
traj_est_vio, None,
"VIO ATE Mapped Onto Trajectory"))
plot_collection.add_figure(
"VIO_RPE_translation",
plot_metric(rpe_metric_trans_vio, "VIO RPE Translation"))
plot_collection.add_figure(
"VIO_RPE_translation_trajectory_error",
plot_traj_colormap_rpe(
rpe_metric_trans_vio, traj_ref_vio, traj_est_vio, None,
"VIO RPE Translation Error Mapped Onto Trajectory"))
plot_collection.add_figure(
"VIO_RPE_Rotation",
plot_metric(rpe_metric_rot_vio, "VIO RPE Rotation"))
plot_collection.add_figure(
"VIO_RPE_rotation_trajectory_error",
plot_traj_colormap_rpe(
rpe_metric_rot_vio, traj_ref_vio, traj_est_vio, None,
"VIO RPE Rotation Error Mapped Onto Trajectory"))
return [plot_collection, results_vio, results_pgo]
# res_final = []
for seq in seqs:
traj_files = glob.glob(str(data_path / f'{seq}/*.txt'))
# print(data_path / f'{seq}')
gt_file = f'{gt_path}/{seq}.csv'
failure_count = 0
if not len(traj_files) == 0:
mean, rmse = [], []
for traj_file in traj_files:
traj_gt = file_interface.read_euroc_csv_trajectory(gt_file)
traj_est = file_interface.read_tum_trajectory_file(traj_file)
kf_traj_est = file_interface.read_tum_trajectory_file(traj_file)
traj_gt, traj_est = sync.associate_trajectories(traj_gt, traj_est)
traj_est_aligned, rot, trans, scale = trajectory.align_trajectory(
traj_est, traj_gt, correct_scale=True, return_parameters=True)
data = (traj_gt, traj_est_aligned)
ape_metric = metrics.APE(pose_relation=pose_relation)
ape_metric.process_data(data)
ape_stat = ape_metric.get_all_statistics()
mean_curr = ape_stat['mean']
rmse_curr = ape_stat['rmse']
if mean_curr > 1.0 or rmse_curr > 1.0:
failure_count += 1
continue
mean.append(mean_curr)
rmse.append(rmse_curr)
print(
def run(args):
import os
import sys
import numpy as np
import evo.core.lie_algebra as lie
from evo.core import trajectory
from evo.core.trajectory import PoseTrajectory3D
from evo.tools import file_interface, log
from evo.tools.settings import SETTINGS
log.configure_logging(verbose=args.verbose,
silent=args.silent,
debug=args.debug)
if args.debug:
import pprint
logger.debug(
"main_parser config:\n" +
pprint.pformat({arg: getattr(args, arg)
for arg in vars(args)}) + "\n")
logger.debug(SEP)
trajectories, ref_traj = load_trajectories(args)
if args.merge:
if args.subcommand == "kitti":
die("Can't merge KITTI files.")
if len(trajectories) == 0:
die("No trajectories to merge (excluding --ref).")
trajectories = {
"merged_trajectory": trajectory.merge(trajectories.values())
}
if args.transform_left or args.transform_right:
tf_type = "left" if args.transform_left else "right"
tf_path = args.transform_left \
if args.transform_left else args.transform_right
transform = file_interface.load_transform_json(tf_path)
logger.debug(SEP)
if not lie.is_se3(transform):
logger.warning("Not a valid SE(3) transformation!")
if args.invert_transform:
transform = lie.se3_inverse(transform)
logger.debug("Applying a {}-multiplicative transformation:\n{}".format(
tf_type, transform))
for traj in trajectories.values():
traj.transform(transform, right_mul=args.transform_right)
if args.t_offset:
logger.debug(SEP)
for name, traj in trajectories.items():
if type(traj) is trajectory.PosePath3D:
die("{} doesn't have timestamps - can't add time offset.".
format(name))
logger.info("Adding time offset to {}: {} (s)".format(
name, args.t_offset))
traj.timestamps += args.t_offset
if args.sync or args.align or args.correct_scale:
from evo.core import sync
if not args.ref:
logger.debug(SEP)
die("Can't align or sync without a reference! (--ref) *grunt*")
for name, traj in trajectories.items():
if args.subcommand == "kitti":
ref_traj_tmp = ref_traj
else:
logger.debug(SEP)
ref_traj_tmp, trajectories[name] = sync.associate_trajectories(
ref_traj,
traj,
max_diff=args.t_max_diff,
first_name="reference",
snd_name=name)
if args.align or args.correct_scale:
logger.debug(SEP)
logger.debug("Aligning {} to reference.".format(name))
trajectories[name] = trajectory.align_trajectory(
trajectories[name],
ref_traj_tmp,
correct_scale=args.correct_scale,
correct_only_scale=args.correct_scale and not args.align,
n=args.n_to_align)
for name, traj in trajectories.items():
print_traj_info(name, traj, args.verbose, args.full_check)
if args.ref:
print_traj_info(args.ref, ref_traj, args.verbose, args.full_check)
if args.plot or args.save_plot or args.serialize_plot:
from evo.tools.plot import PlotMode
plot_mode = PlotMode.xyz if not args.plot_mode else PlotMode[
args.plot_mode]
import numpy as np
from evo.tools import plot
import matplotlib.pyplot as plt
import matplotlib.cm as cm
plot_collection = plot.PlotCollection("evo_traj - trajectory plot")
fig_xyz, axarr_xyz = plt.subplots(3,
sharex="col",
figsize=tuple(SETTINGS.plot_figsize))
fig_rpy, axarr_rpy = plt.subplots(3,
sharex="col",
figsize=tuple(SETTINGS.plot_figsize))
fig_traj = plt.figure(figsize=tuple(SETTINGS.plot_figsize))
ax_traj = plot.prepare_axis(fig_traj, plot_mode)
pltstart = 0
pltend = traj.num_poses
if args.plotstart:
pltstart = args.plotstart
if args.plotend != -1:
pltend = args.plotend
if args.ref:
short_traj_name = os.path.splitext(os.path.basename(args.ref))[0]
if SETTINGS.plot_usetex:
short_traj_name = short_traj_name.replace("_", "\\_")
plot.traj(ax_traj,
plot_mode,
ref_traj,
style=SETTINGS.plot_reference_linestyle,
color=SETTINGS.plot_reference_color,
label=short_traj_name,
alpha=SETTINGS.plot_reference_alpha,
start=pltstart,
end=pltend)
plot.traj_xyz(axarr_xyz,
ref_traj,
style=SETTINGS.plot_reference_linestyle,
color=SETTINGS.plot_reference_color,
label=short_traj_name,
alpha=SETTINGS.plot_reference_alpha,
start=pltstart,
end=pltend)
plot.traj_rpy(axarr_rpy,
ref_traj,
style=SETTINGS.plot_reference_linestyle,
color=SETTINGS.plot_reference_color,
label=short_traj_name,
alpha=SETTINGS.plot_reference_alpha,
start=pltstart,
end=pltend)
cmap_colors = None
if SETTINGS.plot_multi_cmap.lower() != "none":
cmap = getattr(cm, SETTINGS.plot_multi_cmap)
cmap_colors = iter(cmap(np.linspace(0, 1, len(trajectories))))
fig_3 = plt.figure(figsize=tuple(SETTINGS.plot_figsize))
#plot_mode = plot.PlotMode.xz
ax = plot.prepare_axis(fig_3, plot_mode)
fig_3.axes.append(ax)
for name, traj in trajectories.items():
num = traj.positions_xyz.shape[0]
if pltstart >= num:
print(name, "plotstart > len!", num)
pltstart = 0
if pltend != -1 and (pltend > num or pltend < pltstart):
print(name, "plotend > len!", num)
pltend = traj.num_poses
pltstart = int(pltstart)
pltend = int(pltend)
if cmap_colors is None:
color = next(ax_traj._get_lines.prop_cycler)['color']
else:
color = next(cmap_colors)
short_traj_name = os.path.splitext(os.path.basename(name))[0]
if SETTINGS.plot_usetex:
short_traj_name = short_traj_name.replace("_", "\\_")
plot.traj(ax_traj,
plot_mode,
traj,
'-',
color,
short_traj_name,
start=pltstart,
end=pltend)
if args.ref and isinstance(ref_traj, trajectory.PoseTrajectory3D):
start_time = ref_traj.timestamps[0]
else:
start_time = None
plot.traj_xyz(axarr_xyz,
traj,
'-',
color,
short_traj_name,
start_timestamp=start_time,
start=pltstart,
end=pltend)
plot.traj_rpy(axarr_rpy,
traj,
'-',
color,
short_traj_name,
start_timestamp=start_time,
start=pltstart,
end=pltend)
speeds = [
trajectory.calc_speed(traj.positions_xyz[i],
traj.positions_xyz[i + 1],
traj.timestamps[i],
traj.timestamps[i + 1])
for i in range(pltstart, pltend - 1)
]
speeds.append(0)
#plot.traj(ax, plot_mode, traj, '--', 'gray', 'reference')
plot.traj_colormap(ax,
traj,
speeds,
plot_mode,
min_map=min(speeds),
max_map=max(max(speeds), 10),
title="speed mapped onto trajectory",
start=pltstart,
end=pltend)
plot_collection.add_figure("trajectories", fig_traj)
plot_collection.add_figure("xyz_view", fig_xyz)
plot_collection.add_figure("rpy_view", fig_rpy)
plot_collection.add_figure("traj (speed)", fig_3)
if args.plot:
plot_collection.show()
if args.save_plot:
logger.info(SEP)
plot_collection.export(args.save_plot,
confirm_overwrite=not args.no_warnings)
if args.serialize_plot:
logger.info(SEP)
plot_collection.serialize(args.serialize_plot,
confirm_overwrite=not args.no_warnings)
if args.save_as_tum:
logger.info(SEP)
for name, traj in trajectories.items():
dest = os.path.splitext(os.path.basename(name))[0] + ".tum"
file_interface.write_tum_trajectory_file(
dest, traj, confirm_overwrite=not args.no_warnings)
if args.ref:
dest = os.path.splitext(os.path.basename(args.ref))[0] + ".tum"
file_interface.write_tum_trajectory_file(
dest, ref_traj, confirm_overwrite=not args.no_warnings)
if args.save_as_kitti:
logger.info(SEP)
for name, traj in trajectories.items():
dest = os.path.splitext(os.path.basename(name))[0] + ".kitti"
file_interface.write_kitti_poses_file(
dest, traj, confirm_overwrite=not args.no_warnings)
if args.ref:
dest = os.path.splitext(os.path.basename(args.ref))[0] + ".kitti"
file_interface.write_kitti_poses_file(
dest, ref_traj, confirm_overwrite=not args.no_warnings)
if args.save_as_bag:
import datetime
import rosbag
dest_bag_path = str(
datetime.datetime.now().strftime("%Y-%m-%d-%H-%M-%S")) + ".bag"
logger.info(SEP)
logger.info("Saving trajectories to " + dest_bag_path + "...")
bag = rosbag.Bag(dest_bag_path, 'w')
try:
for name, traj in trajectories.items():
dest_topic = os.path.splitext(os.path.basename(name))[0]
frame_id = traj.meta[
"frame_id"] if "frame_id" in traj.meta else ""
file_interface.write_bag_trajectory(bag, traj, dest_topic,
frame_id)
if args.ref:
dest_topic = os.path.splitext(os.path.basename(args.ref))[0]
frame_id = ref_traj.meta[
"frame_id"] if "frame_id" in ref_traj.meta else ""
file_interface.write_bag_trajectory(bag, ref_traj, dest_topic,
frame_id)
finally:
bag.close()
import numpy as np
import matplotlib.pyplot as plt
logging.basicConfig(format="%(message)s",
stream=sys.stdout,
level=logging.DEBUG)
traj_ref = file_interface.read_kitti_poses_file("../data/KITTI_00_gt.txt")
traj_est = file_interface.read_kitti_poses_file("../data/KITTI_00_ORB.txt")
# add artificial Sim(3) transformation
traj_est.transform(lie.se3(np.eye(3), [0, 0, 0]))
traj_est.scale(0.5)
logging.info("\nUmeyama alignment without scaling")
traj_est_aligned = trajectory.align_trajectory(traj_est, traj_ref)
logging.info("\nUmeyama alignment with scaling")
traj_est_aligned_scaled = trajectory.align_trajectory(traj_est,
traj_ref,
correct_scale=True)
logging.info("\nUmeyama alignment with scaling only")
traj_est_aligned_only_scaled = trajectory.align_trajectory(
traj_est, traj_ref, correct_only_scale=True)
fig = plt.figure(figsize=(8, 8))
# fig.suptitle("Umeyama $\mathrm{Sim}(3)$ alignment")
plot_mode = plot.PlotMode.xz
ax = plot.prepare_axis(fig, plot_mode,
subplot_arg='221') # 122 = 1*2 grid, second
plot.traj(ax, plot_mode, traj_ref, '--', 'gray') #, 'reference')
plot.traj(ax, plot_mode, traj_est, '-', 'blue') #, 'not aligned')
def main_rpe_for_each(traj_ref, traj_est, pose_relation, mode, bins, rel_tols,
align=False, correct_scale=False, ref_name="", est_name="",
show_plot=False, save_plot=None, save_results=None, no_warnings=False,
serialize_plot=None):
from evo.core import metrics, result
from evo.core import filters
from evo.core import trajectory
from evo.tools import file_interface
from evo.tools.settings import SETTINGS
if not bins or not rel_tols:
raise RuntimeError("bins and tolerances must have more than one element")
if len(bins) != len(rel_tols):
raise RuntimeError("bins and tolerances must have the same number of elements")
if mode in {"speed", "angular_speed"} and traj_est is trajectory.PosePath3D:
raise RuntimeError("timestamps are required for mode: " + mode)
bin_unit = None
if mode == "speed":
bin_unit = metrics.VelUnit.meters_per_sec
elif mode == "path":
bin_unit = metrics.Unit.meters
elif mode == "angle":
bin_unit = metrics.Unit.degrees
elif mode == "angular_speed":
bin_unit = metrics.VelUnit.degrees_per_sec
rpe_unit = None
if pose_relation is metrics.PoseRelation.translation_part:
rpe_unit = metrics.Unit.meters
elif pose_relation is metrics.PoseRelation.rotation_angle_deg:
rpe_unit = metrics.Unit.degrees
elif pose_relation is metrics.PoseRelation.rotation_angle_rad:
rpe_unit = metrics.Unit.radians
correct_only_scale = correct_scale and not align
if align or correct_scale:
logger.debug(SEP)
if correct_only_scale:
logger.debug("Correcting scale...")
else:
logger.debug("Aligning using Umeyama's method..."
+ (" (with scale correction)" if correct_scale else ""))
traj_est = trajectory.align_trajectory(traj_est, traj_ref, correct_scale,
correct_only_scale)
results = []
for bin, rel_tol, in zip(bins, rel_tols):
logger.debug(SEP)
logger.info(
"Calculating RPE for each sub-sequence of " + str(bin) + " (" + bin_unit.value + ")")
tol = bin * rel_tol
id_pairs = []
if mode == "path":
id_pairs = filters.filter_pairs_by_path(traj_ref.poses_se3, bin, tol, all_pairs=True)
elif mode == "angle":
id_pairs = filters.filter_pairs_by_angle(traj_ref.poses_se3, bin, tol, degrees=True)
elif mode == "speed":
id_pairs = filters.filter_pairs_by_speed(traj_ref.poses_se3, traj_ref.timestamps,
bin, tol)
elif mode == "angular_speed":
id_pairs = filters.filter_pairs_by_angular_speed(traj_ref.poses_se3,
traj_ref.timestamps, bin, tol, True)
if len(id_pairs) == 0:
raise RuntimeError("bin " + str(bin) + " (" + str(bin_unit.value) + ") "
+ "produced empty index list - try other values")
# calculate RPE with all IDs (delta 1 frames)
data = (traj_ref, traj_est)
# the delta here has nothing to do with the bin - 1f delta just to use all poses of the bin
rpe_metric = metrics.RPE(pose_relation, delta=1, delta_unit=metrics.Unit.frames,
all_pairs=True)
rpe_metric.process_data(data, id_pairs)
mean = rpe_metric.get_statistic(metrics.StatisticsType.mean)
results.append(mean)
if SETTINGS.plot_usetex:
mode.replace("_", "\_")
title = "mean RPE w.r.t. " + pose_relation.value + "\nfor different " + mode + " sub-sequences"
if align and not correct_scale:
title += "\n(with SE(3) Umeyama alignment)"
elif align and correct_scale:
title += "\n(with Sim(3) Umeyama alignment)"
elif correct_only_scale:
title += "\n(scale corrected)"
else:
title += "\n(not aligned)"
rpe_for_each_result = result.Result()
rpe_for_each_result.add_info({
"label": "RPE ({})".format(rpe_unit.value),
"est_name": est_name,
"ref_name": ref_name,
"title": title,
"xlabel": "{} sub-sequences ({})".format(mode, bin_unit.value)
})
rpe_for_each_result.add_stats({bin: result for bin, result in zip(bins, results)})
# TODO use a more suitable name than seconds
rpe_for_each_result.add_np_array("seconds_from_start", bins)
rpe_for_each_result.add_np_array("error_array", results)
logger.debug(SEP)
logger.info(rpe_for_each_result.pretty_str())
if show_plot or save_plot or serialize_plot:
from evo.tools import plot
import matplotlib.pyplot as plt
plot_collection = plot.PlotCollection(title)
fig = plt.figure(figsize=(SETTINGS.plot_figsize[0], SETTINGS.plot_figsize[1]))
plot.error_array(fig, results, x_array=bins,
name="mean RPE" + (" (" + rpe_unit.value + ")") if rpe_unit else "",
marker="o", title=title,
xlabel=mode + " sub-sequences " + " (" + bin_unit.value + ")")
# info text
if SETTINGS.plot_info_text and est_name and ref_name:
ax = fig.gca()
ax.text(0, -0.12, "estimate: " + est_name + "\nreference: " + ref_name,
transform=ax.transAxes, fontsize=8, color="gray")
plt.title(title)
plot_collection.add_figure("raw", fig)
if show_plot:
plot_collection.show()
if save_plot:
plot_collection.export(save_plot, confirm_overwrite=not no_warnings)
if serialize_plot:
logger.debug(SEP)
plot_collection.serialize(serialize_plot, confirm_overwrite=not no_warnings)
if save_results:
logger.debug(SEP)
if SETTINGS.save_traj_in_zip:
rpe_for_each_result.add_trajectory("traj_ref", traj_ref)
rpe_for_each_result.add_trajectory("traj_est", traj_est)
file_interface.save_res_file(save_results, rpe_for_each_result,
confirm_overwrite=not no_warnings)
return rpe_for_each_result
def main_rpe(traj_ref,
traj_est,
pose_relation,
delta,
delta_unit,
rel_delta_tol=0.1,
all_pairs=False,
align=False,
correct_scale=False,
ref_name="",
est_name="",
show_plot=False,
save_plot=None,
plot_mode=None,
save_results=None,
no_warnings=False,
support_loop=False,
serialize_plot=None):
from evo.core import metrics, result
from evo.core import trajectory
from evo.tools import file_interface
from evo.tools.settings import SETTINGS
if (show_plot or save_plot or serialize_plot) and all_pairs:
raise metrics.MetricsException(
"all_pairs mode cannot be used with plotting functions")
only_scale = correct_scale and not align
if align or correct_scale:
logging.debug(SEP)
if only_scale:
logging.debug("correcting scale...")
else:
logging.debug("aligning using Umeyama's method..." + (
" (with scale correction)" if correct_scale else ""))
traj_est = trajectory.align_trajectory(traj_est, traj_ref,
correct_scale, only_scale)
logging.debug(SEP)
# calculate RPE
data = (traj_ref, traj_est)
rpe_metric = metrics.RPE(pose_relation, delta, delta_unit, rel_delta_tol,
all_pairs)
rpe_metric.process_data(data)
rpe_statistics = rpe_metric.get_all_statistics()
title = str(rpe_metric)
if align and not correct_scale:
title += "\n(with SE(3) Umeyama alignment)"
elif align and correct_scale:
title += "\n(with Sim(3) Umeyama alignment)"
elif only_scale:
title += "\n(scale corrected)"
else:
title += "\n(not aligned)"
rpe_result = result.from_metric(rpe_metric, title, ref_name, est_name)
logging.debug(SEP)
logging.info(rpe_result.pretty_str())
# restrict trajectories to delta ids
if support_loop:
# avoid overwriting if called repeatedly e.g. in Jupyter notebook
import copy
traj_ref = copy.deepcopy(traj_ref)
traj_est = copy.deepcopy(traj_est)
traj_ref.reduce_to_ids(rpe_metric.delta_ids)
traj_est.reduce_to_ids(rpe_metric.delta_ids)
if isinstance(traj_est, trajectory.PoseTrajectory3D) and not all_pairs:
seconds_from_start = [
t - traj_est.timestamps[0] for t in traj_est.timestamps
]
rpe_result.add_np_array("seconds_from_start", seconds_from_start)
else:
seconds_from_start = None
if show_plot or save_plot or serialize_plot and not all_pairs:
from evo.tools import plot
import matplotlib.pyplot as plt
logging.debug(SEP)
logging.debug("plotting results... ")
fig1 = plt.figure(figsize=(SETTINGS.plot_figsize[0],
SETTINGS.plot_figsize[1]))
# metric values
plot.error_array(
fig1,
rpe_metric.error,
x_array=seconds_from_start,
statistics=rpe_statistics,
name="RPE" +
(" (" + rpe_metric.unit.value + ")") if rpe_metric.unit else "",
title=title,
xlabel="$t$ (s)" if seconds_from_start else "index")
# info text
if SETTINGS.plot_info_text and est_name and ref_name:
ax = fig1.gca()
ax.text(0,
-0.12,
"estimate: " + est_name + "\nreference: " + ref_name,
transform=ax.transAxes,
fontsize=8,
color="gray")
# trajectory colormapped
fig2 = plt.figure(figsize=(SETTINGS.plot_figsize[0],
SETTINGS.plot_figsize[1]))
plot_mode = plot_mode if plot_mode is not None else plot.PlotMode.xyz
ax = plot.prepare_axis(fig2, plot_mode)
plot.traj(ax,
plot_mode,
traj_ref,
'--',
'gray',
'reference',
alpha=0 if SETTINGS.plot_hideref else 1)
plot.traj_colormap(ax,
traj_est,
rpe_metric.error,
plot_mode,
min_map=rpe_statistics["min"],
max_map=rpe_statistics["max"],
title="RPE mapped onto trajectory")
fig2.axes.append(ax)
plot_collection = plot.PlotCollection(title)
plot_collection.add_figure("raw", fig1)
plot_collection.add_figure("map", fig2)
if show_plot:
plot_collection.show()
if save_plot:
plot_collection.export(save_plot,
confirm_overwrite=not no_warnings)
if serialize_plot:
logging.debug(SEP)
plot_collection.serialize(serialize_plot,
confirm_overwrite=not no_warnings)
if save_results:
logging.debug(SEP)
if SETTINGS.save_traj_in_zip:
rpe_result.add_trajectory("traj_ref", traj_ref)
rpe_result.add_trajectory("traj_est", traj_est)
file_interface.save_res_file(save_results,
rpe_result,
confirm_overwrite=not no_warnings)
return rpe_result
def run(args):
import os
import sys
import numpy as np
import evo.core.lie_algebra as lie
from evo.core import trajectory
from evo.core.trajectory import PoseTrajectory3D
from evo.tools import file_interface, log
from evo.tools.settings import SETTINGS
log.configure_logging(verbose=args.verbose,
silent=args.silent,
debug=args.debug)
if args.debug:
import pprint
logger.debug(
"main_parser config:\n" +
pprint.pformat({arg: getattr(args, arg)
for arg in vars(args)}) + "\n")
logger.debug(SEP)
trajectories = []
ref_traj = None
if args.subcommand == "tum":
for traj_file in args.traj_files:
if traj_file != args.ref:
trajectories.append(
(traj_file,
file_interface.read_tum_trajectory_file(traj_file)))
if args.ref:
ref_traj = file_interface.read_tum_trajectory_file(args.ref)
elif args.subcommand == "kitti":
for pose_file in args.pose_files:
if pose_file != args.ref:
trajectories.append(
(pose_file,
file_interface.read_kitti_poses_file(pose_file)))
if args.ref:
ref_traj = file_interface.read_kitti_poses_file(args.ref)
elif args.subcommand == "euroc":
for csv_file in args.state_gt_csv:
if csv_file != args.ref:
trajectories.append(
(csv_file,
file_interface.read_euroc_csv_trajectory(csv_file)))
if args.ref:
ref_traj = file_interface.read_euroc_csv_trajectory(args.ref)
elif args.subcommand == "bag":
import rosbag
bag = rosbag.Bag(args.bag)
try:
if args.all_topics:
topic_info = bag.get_type_and_topic_info()
topics = sorted([
t for t in topic_info[1].keys()
if topic_info[1][t][0] == "geometry_msgs/PoseStamped"
and t != args.ref
])
if len(topics) == 0:
logger.error("No geometry_msgs/PoseStamped topics found!")
sys.exit(1)
else:
topics = args.topics
if not topics:
logger.warning(
"No topics used - specify topics or set --all_topics.")
sys.exit(1)
for topic in topics:
trajectories.append(
(topic, file_interface.read_bag_trajectory(bag, topic)))
if args.ref:
ref_traj = file_interface.read_bag_trajectory(bag, args.ref)
finally:
bag.close()
else:
raise RuntimeError("unsupported subcommand: " + args.subcommand)
if args.merge:
if args.subcommand == "kitti":
raise TypeError(
"can't merge KITTI files - but you can append them with 'cat'")
if len(trajectories) == 0:
raise RuntimeError("no trajectories to merge (excluding --ref)")
merged_stamps = trajectories[0][1].timestamps
merged_xyz = trajectories[0][1].positions_xyz
merged_quat = trajectories[0][1].orientations_quat_wxyz
for _, traj in trajectories[1:]:
merged_stamps = np.concatenate((merged_stamps, traj.timestamps))
merged_xyz = np.concatenate((merged_xyz, traj.positions_xyz))
merged_quat = np.concatenate(
(merged_quat, traj.orientations_quat_wxyz))
order = merged_stamps.argsort()
merged_stamps = merged_stamps[order]
merged_xyz = merged_xyz[order]
merged_quat = merged_quat[order]
trajectories = [("merged_trajectory",
PoseTrajectory3D(merged_xyz, merged_quat,
merged_stamps))]
if args.transform_left or args.transform_right:
tf_type = "left" if args.transform_left else "right"
tf_path = args.transform_left if args.transform_left else args.transform_right
t, xyz, quat = file_interface.load_transform_json(tf_path)
logger.debug(SEP)
if not lie.is_se3(t):
logger.warning("Not a valid SE(3) transformation!")
if args.invert_transform:
t = lie.se3_inverse(t)
logger.debug("Applying a {}-multiplicative transformation:\n{}".format(
tf_type, t))
for name, traj in trajectories:
traj.transform(t, right_mul=args.transform_right)
if args.t_offset:
logger.debug(SEP)
for name, traj in trajectories:
if type(traj) is trajectory.PosePath3D:
logger.warning(
"{} doesn't have timestamps - can't add time offset.".
format(name))
else:
logger.info("Adding time offset to {}: {} (s)".format(
name, args.t_offset))
traj.timestamps += args.t_offset
if args.align or args.correct_scale:
if not args.ref:
logger.debug(SEP)
logger.warning("Can't align without a reference! (--ref) *grunt*")
else:
if args.subcommand == "kitti":
traj_tmp, ref_traj_tmp = trajectories, [
ref_traj for n, t in trajectories
]
else:
traj_tmp, ref_traj_tmp = [], []
from evo.core import sync
for name, traj in trajectories:
logger.debug(SEP)
ref_assoc, traj_assoc = sync.associate_trajectories(
ref_traj,
traj,
max_diff=args.t_max_diff,
first_name="ref",
snd_name=name)
ref_traj_tmp.append(ref_assoc)
traj_tmp.append((name, traj_assoc))
trajectories = traj_tmp
correct_only_scale = args.correct_scale and not args.align
trajectories_new = []
for nt, ref_assoc in zip(trajectories, ref_traj_tmp):
logger.debug(SEP)
logger.debug("Aligning " + nt[0] + " to " + args.ref + "...")
trajectories_new.append(
(nt[0],
trajectory.align_trajectory(nt[1], ref_assoc,
args.correct_scale,
correct_only_scale,
args.n_to_align)))
trajectories = trajectories_new
for name, traj in trajectories:
print_traj_info(name, traj, args.verbose, args.full_check)
if args.ref:
print_traj_info(args.ref, ref_traj, args.verbose, args.full_check)
if args.plot or args.save_plot or args.serialize_plot:
from evo.tools.plot import PlotMode
plot_mode = PlotMode.xyz if not args.plot_mode else PlotMode[
args.plot_mode]
import numpy as np
from evo.tools import plot
import matplotlib.pyplot as plt
import matplotlib.cm as cm
plot_collection = plot.PlotCollection("evo_traj - trajectory plot")
fig_xyz, axarr_xyz = plt.subplots(3,
sharex="col",
figsize=tuple(SETTINGS.plot_figsize))
fig_rpy, axarr_rpy = plt.subplots(3,
sharex="col",
figsize=tuple(SETTINGS.plot_figsize))
fig_traj = plt.figure(figsize=tuple(SETTINGS.plot_figsize))
if (args.align or args.correct_scale) and not args.ref:
plt.xkcd(scale=2, randomness=4)
fig_traj.suptitle("what if --ref?")
fig_xyz.suptitle("what if --ref?")
ax_traj = plot.prepare_axis(fig_traj, plot_mode)
if args.ref:
short_traj_name = os.path.splitext(os.path.basename(args.ref))[0]
if SETTINGS.plot_usetex:
short_traj_name = short_traj_name.replace("_", "\\_")
plot.traj(ax_traj,
plot_mode,
ref_traj,
'--',
'grey',
short_traj_name,
alpha=0 if SETTINGS.plot_hideref else 1)
plot.traj_xyz(axarr_xyz,
ref_traj,
'--',
'grey',
short_traj_name,
alpha=0 if SETTINGS.plot_hideref else 1)
plot.traj_rpy(axarr_rpy,
ref_traj,
'--',
'grey',
short_traj_name,
alpha=0 if SETTINGS.plot_hideref else 1)
cmap_colors = None
if SETTINGS.plot_multi_cmap.lower() != "none":
cmap = getattr(cm, SETTINGS.plot_multi_cmap)
cmap_colors = iter(cmap(np.linspace(0, 1, len(trajectories))))
for name, traj in trajectories:
if cmap_colors is None:
color = next(ax_traj._get_lines.prop_cycler)['color']
else:
color = next(cmap_colors)
short_traj_name = os.path.splitext(os.path.basename(name))[0]
if SETTINGS.plot_usetex:
short_traj_name = short_traj_name.replace("_", "\\_")
plot.traj(ax_traj, plot_mode, traj, '-', color, short_traj_name)
if args.ref and isinstance(ref_traj, trajectory.PoseTrajectory3D):
start_time = ref_traj.timestamps[0]
else:
start_time = None
plot.traj_xyz(axarr_xyz,
traj,
'-',
color,
short_traj_name,
start_timestamp=start_time)
plot.traj_rpy(axarr_rpy,
traj,
'-',
color,
short_traj_name,
start_timestamp=start_time)
plt.tight_layout()
plot_collection.add_figure("trajectories", fig_traj)
plot_collection.add_figure("xyz_view", fig_xyz)
plot_collection.add_figure("rpy_view", fig_rpy)
if args.plot:
plot_collection.show()
if args.save_plot:
logger.info(SEP)
plot_collection.export(args.save_plot,
confirm_overwrite=not args.no_warnings)
if args.serialize_plot:
logger.info(SEP)
plot_collection.serialize(args.serialize_plot,
confirm_overwrite=not args.no_warnings)
if args.save_as_tum:
logger.info(SEP)
for name, traj in trajectories:
dest = os.path.splitext(os.path.basename(name))[0] + ".tum"
file_interface.write_tum_trajectory_file(
dest, traj, confirm_overwrite=not args.no_warnings)
if args.ref:
dest = os.path.splitext(os.path.basename(args.ref))[0] + ".tum"
file_interface.write_tum_trajectory_file(
dest, ref_traj, confirm_overwrite=not args.no_warnings)
if args.save_as_kitti:
logger.info(SEP)
for name, traj in trajectories:
dest = os.path.splitext(os.path.basename(name))[0] + ".kitti"
file_interface.write_kitti_poses_file(
dest, traj, confirm_overwrite=not args.no_warnings)
if args.ref:
dest = os.path.splitext(os.path.basename(args.ref))[0] + ".kitti"
file_interface.write_kitti_poses_file(
dest, ref_traj, confirm_overwrite=not args.no_warnings)
if args.save_as_bag:
logger.info(SEP)
import datetime
import rosbag
dest_bag_path = str(
datetime.datetime.now().strftime('%Y-%m-%d_%H:%M:%S.%f')) + ".bag"
logger.info("Saving trajectories to " + dest_bag_path + "...")
bag = rosbag.Bag(dest_bag_path, 'w')
try:
for name, traj in trajectories:
dest_topic = os.path.splitext(os.path.basename(name))[0]
frame_id = traj.meta[
"frame_id"] if "frame_id" in traj.meta else ""
file_interface.write_bag_trajectory(bag, traj, dest_topic,
frame_id)
if args.ref:
dest_topic = os.path.splitext(os.path.basename(args.ref))[0]
frame_id = ref_traj.meta[
"frame_id"] if "frame_id" in ref_traj.meta else ""
file_interface.write_bag_trajectory(bag, ref_traj, dest_topic,
frame_id)
finally:
bag.close()
# %%
# Convert the gt relative-pose DataFrame to a trajectory object.
traj_ref_complete = pandas_bridge.df_to_trajectory(gt_df)
# Use the backend poses as trajectory.
traj_est_unaligned = pandas_bridge.df_to_trajectory(output_poses_df)
discard_n_start_poses = 0
discard_n_end_poses = 0
# Associate the data.
traj_est = copy.deepcopy(traj_est_unaligned)
traj_ref, traj_est = sync.associate_trajectories(traj_ref_complete, traj_est)
traj_est = trajectory.align_trajectory(
traj_est,
traj_ref,
correct_scale=False,
discard_n_start_poses=int(discard_n_start_poses),
discard_n_end_poses=int(discard_n_end_poses),
)
print("traj_ref: ", traj_ref)
print("traj_est: ", traj_est)
# %%
# plot ground truth trajectory with pose
plot_mode = plot.PlotMode.xy
fig = plt.figure()
ax = plot.prepare_axis(fig, plot_mode)
draw_coordinate_axes(ax,
traj_ref_complete,
marker_scale=2,
vanilla_traj = trajectory.PoseTrajectory3D(poses_se3=vanilla_poses,
timestamps=timestamps)
vision_only_traj = trajectory.PoseTrajectory3D(poses_se3=vision_only_poses,
timestamps=timestamps)
vins_mono_traj = read_tum(vins_mono_poses)
gt_traj_synced_vanilla, vanilla_traj_synced = sync.associate_trajectories(
gt_traj, vanilla_traj, max_diff=0.01)
gt_traj_synced_vision_only, vision_only_traj_synced = sync.associate_trajectories(
gt_traj, vision_only_traj, max_diff=0.01)
gt_traj_synced_vins_mono, vins_mono_traj_synced = sync.associate_trajectories(
gt_traj, vins_mono_traj, max_diff=0.01)
vanilla_traj_aligned = trajectory.align_trajectory(vanilla_traj_synced,
gt_traj_synced_vanilla,
correct_scale=False,
correct_only_scale=False)
vision_only_traj_aligned = trajectory.align_trajectory(
vision_only_traj_synced,
gt_traj_synced_vision_only,
correct_scale=False,
correct_only_scale=False)
vins_mono_traj_aligned = trajectory.align_trajectory(vins_mono_traj_synced,
gt_traj_synced_vins_mono,
correct_scale=False,
correct_only_scale=False)
plotter = Plotter(os.path.join(base_dir, "KITTI_figures"))
def plot_callback(fig, ax):
from __future__ import print_function
print("loading required evo modules")
from evo.core import trajectory, sync, metrics
from evo.tools import file_interface
print("loading trajectories")
traj_ref = file_interface.read_tum_trajectory_file(
"../../test/data/fr2_desk_groundtruth.txt")
traj_est = file_interface.read_tum_trajectory_file(
"../../test/data/fr2_desk_ORB.txt")
print("registering and aligning trajectories")
traj_ref, traj_est = sync.associate_trajectories(traj_ref, traj_est)
traj_est = trajectory.align_trajectory(traj_est, traj_ref, correct_scale=False)
print("calculating APE")
data = (traj_ref, traj_est)
ape_metric = metrics.APE(metrics.PoseRelation.translation_part)
ape_metric.process_data(data)
ape_statistics = ape_metric.get_all_statistics()
print("mean:", ape_statistics["mean"])
print("loading plot modules")
from evo.tools import plot
import matplotlib.pyplot as plt
print("plotting")
plot_collection = plot.PlotCollection("Example")
# metric values
def run(args):
import os
import sys
import numpy as np
import evo.core.lie_algebra as lie
from evo.core import trajectory
from evo.core.trajectory import PoseTrajectory3D
from evo.tools import file_interface, log
from evo.tools.settings import SETTINGS
log.configure_logging(verbose=args.verbose,
silent=args.silent,
debug=args.debug,
local_logfile=args.logfile)
if args.debug:
import pprint
logger.debug(
"main_parser config:\n" +
pprint.pformat({arg: getattr(args, arg)
for arg in vars(args)}) + "\n")
logger.debug(SEP)
trajectories, ref_traj = load_trajectories(args)
if args.merge:
if args.subcommand == "kitti":
die("Can't merge KITTI files.")
if len(trajectories) == 0:
die("No trajectories to merge (excluding --ref).")
trajectories = {
"merged_trajectory": trajectory.merge(trajectories.values())
}
if args.transform_left or args.transform_right:
tf_type = "left" if args.transform_left else "right"
tf_path = args.transform_left \
if args.transform_left else args.transform_right
transform = file_interface.load_transform_json(tf_path)
logger.debug(SEP)
if not lie.is_se3(transform):
logger.warning("Not a valid SE(3) transformation!")
if args.invert_transform:
transform = lie.se3_inverse(transform)
logger.debug("Applying a {}-multiplicative transformation:\n{}".format(
tf_type, transform))
for traj in trajectories.values():
traj.transform(transform,
right_mul=args.transform_right,
propagate=args.propagate_transform)
if args.t_offset:
logger.debug(SEP)
for name, traj in trajectories.items():
if type(traj) is trajectory.PosePath3D:
die("{} doesn't have timestamps - can't add time offset.".
format(name))
logger.info("Adding time offset to {}: {} (s)".format(
name, args.t_offset))
traj.timestamps += args.t_offset
if args.n_to_align != -1 and not (args.align or args.correct_scale):
die("--n_to_align is useless without --align or/and --correct_scale")
if args.sync or args.align or args.correct_scale or args.align_origin:
from evo.core import sync
if not args.ref:
logger.debug(SEP)
die("Can't align or sync without a reference! (--ref) *grunt*")
for name, traj in trajectories.items():
if args.subcommand == "kitti":
ref_traj_tmp = ref_traj
else:
logger.debug(SEP)
ref_traj_tmp, trajectories[name] = sync.associate_trajectories(
ref_traj,
traj,
max_diff=args.t_max_diff,
first_name="reference",
snd_name=name)
if args.align or args.correct_scale:
logger.debug(SEP)
logger.debug("Aligning {} to reference.".format(name))
trajectories[name] = trajectory.align_trajectory(
trajectories[name],
ref_traj_tmp,
correct_scale=args.correct_scale,
correct_only_scale=args.correct_scale and not args.align,
n=args.n_to_align)
if args.align_origin:
logger.debug(SEP)
logger.debug("Aligning {}'s origin to reference.".format(name))
trajectories[name] = trajectory.align_trajectory_origin(
trajectories[name], ref_traj_tmp)
print_compact_name = not args.subcommand == "bag"
for name, traj in trajectories.items():
print_traj_info(name, traj, args.verbose, args.full_check,
print_compact_name)
if args.ref:
print_traj_info(args.ref, ref_traj, args.verbose, args.full_check,
print_compact_name)
if args.plot or args.save_plot or args.serialize_plot:
import numpy as np
from evo.tools import plot
import matplotlib.pyplot as plt
import matplotlib.cm as cm
plot_collection = plot.PlotCollection("evo_traj - trajectory plot")
fig_xyz, axarr_xyz = plt.subplots(3,
sharex="col",
figsize=tuple(SETTINGS.plot_figsize))
fig_rpy, axarr_rpy = plt.subplots(3,
sharex="col",
figsize=tuple(SETTINGS.plot_figsize))
fig_traj = plt.figure(figsize=tuple(SETTINGS.plot_figsize))
plot_mode = plot.PlotMode[args.plot_mode]
ax_traj = plot.prepare_axis(fig_traj, plot_mode)
if args.ref:
short_traj_name = os.path.splitext(os.path.basename(args.ref))[0]
if SETTINGS.plot_usetex:
short_traj_name = short_traj_name.replace("_", "\\_")
plot.traj(ax_traj,
plot_mode,
ref_traj,
style=SETTINGS.plot_reference_linestyle,
color=SETTINGS.plot_reference_color,
label=short_traj_name,
alpha=SETTINGS.plot_reference_alpha)
plot.draw_coordinate_axes(ax_traj, ref_traj, plot_mode,
SETTINGS.plot_axis_marker_scale)
plot.traj_xyz(axarr_xyz,
ref_traj,
style=SETTINGS.plot_reference_linestyle,
color=SETTINGS.plot_reference_color,
label=short_traj_name,
alpha=SETTINGS.plot_reference_alpha)
plot.traj_rpy(axarr_rpy,
ref_traj,
style=SETTINGS.plot_reference_linestyle,
color=SETTINGS.plot_reference_color,
label=short_traj_name,
alpha=SETTINGS.plot_reference_alpha)
if args.ros_map_yaml:
plot.ros_map(ax_traj, args.ros_map_yaml, plot_mode)
cmap_colors = None
if SETTINGS.plot_multi_cmap.lower() != "none":
cmap = getattr(cm, SETTINGS.plot_multi_cmap)
cmap_colors = iter(cmap(np.linspace(0, 1, len(trajectories))))
for name, traj in trajectories.items():
if cmap_colors is None:
color = next(ax_traj._get_lines.prop_cycler)['color']
else:
color = next(cmap_colors)
if print_compact_name:
short_traj_name = os.path.splitext(os.path.basename(name))[0]
else:
short_traj_name = name
if SETTINGS.plot_usetex:
short_traj_name = short_traj_name.replace("_", "\\_")
plot.traj(ax_traj,
plot_mode,
traj,
SETTINGS.plot_trajectory_linestyle,
color,
short_traj_name,
alpha=SETTINGS.plot_trajectory_alpha)
plot.draw_coordinate_axes(ax_traj, traj, plot_mode,
SETTINGS.plot_axis_marker_scale)
if args.ref and isinstance(ref_traj, trajectory.PoseTrajectory3D):
start_time = ref_traj.timestamps[0]
else:
start_time = None
plot.traj_xyz(axarr_xyz,
traj,
SETTINGS.plot_trajectory_linestyle,
color,
short_traj_name,
alpha=SETTINGS.plot_trajectory_alpha,
start_timestamp=start_time)
plot.traj_rpy(axarr_rpy,
traj,
SETTINGS.plot_trajectory_linestyle,
color,
short_traj_name,
alpha=SETTINGS.plot_trajectory_alpha,
start_timestamp=start_time)
if not SETTINGS.plot_usetex:
fig_rpy.text(0.,
0.005,
"euler_angle_sequence: {}".format(
SETTINGS.euler_angle_sequence),
fontsize=6)
plot_collection.add_figure("trajectories", fig_traj)
plot_collection.add_figure("xyz_view", fig_xyz)
plot_collection.add_figure("rpy_view", fig_rpy)
if args.plot:
plot_collection.show()
if args.save_plot:
logger.info(SEP)
plot_collection.export(args.save_plot,
confirm_overwrite=not args.no_warnings)
if args.serialize_plot:
logger.info(SEP)
plot_collection.serialize(args.serialize_plot,
confirm_overwrite=not args.no_warnings)
if args.save_as_tum:
logger.info(SEP)
for name, traj in trajectories.items():
dest = os.path.splitext(os.path.basename(name))[0] + ".tum"
file_interface.write_tum_trajectory_file(
dest, traj, confirm_overwrite=not args.no_warnings)
if args.ref:
dest = os.path.splitext(os.path.basename(args.ref))[0] + ".tum"
file_interface.write_tum_trajectory_file(
dest, ref_traj, confirm_overwrite=not args.no_warnings)
if args.save_as_kitti:
logger.info(SEP)
for name, traj in trajectories.items():
dest = os.path.splitext(os.path.basename(name))[0] + ".kitti"
file_interface.write_kitti_poses_file(
dest, traj, confirm_overwrite=not args.no_warnings)
if args.ref:
dest = os.path.splitext(os.path.basename(args.ref))[0] + ".kitti"
file_interface.write_kitti_poses_file(
dest, ref_traj, confirm_overwrite=not args.no_warnings)
if args.save_as_bag:
import datetime
import rosbag
dest_bag_path = str(
datetime.datetime.now().strftime("%Y-%m-%d-%H-%M-%S")) + ".bag"
logger.info(SEP)
logger.info("Saving trajectories to " + dest_bag_path + "...")
bag = rosbag.Bag(dest_bag_path, 'w')
try:
for name, traj in trajectories.items():
dest_topic = os.path.splitext(os.path.basename(name))[0]
frame_id = traj.meta[
"frame_id"] if "frame_id" in traj.meta else ""
file_interface.write_bag_trajectory(bag, traj, dest_topic,
frame_id)
if args.ref:
dest_topic = os.path.splitext(os.path.basename(args.ref))[0]
frame_id = ref_traj.meta[
"frame_id"] if "frame_id" in ref_traj.meta else ""
file_interface.write_bag_trajectory(bag, ref_traj, dest_topic,
frame_id)
finally:
bag.close()
def associate_segments(traj, tracks):
"""Associate segments of an object trajectory as given by a DATMO system
with the object's reference trajectory
:traj: Reference trajectory
:tracks: All the tracks that got produced by the DATMO system
:localization: The trajectory of the self-localization
:returns: segments: The tracks that match to the reference trajectory
:returns: traj_ref: The part of the reference trajectory that matches with
tracks
"""
matches = []
for tr in tracks: # Find the best matching tracks to the object trajectory
traj_ref, traj_est = sync.associate_trajectories(traj,
tr,
max_diff=0.01)
traj_est, rot, tra, _ = trajectory.align_trajectory(
traj_est, traj_ref, correct_scale=False, return_parameters=True)
# print("calculating APE for track of length", len(tr.timestamps))
data = (traj_ref, traj_est)
ape_metric = metrics.APE(metrics.PoseRelation.translation_part)
ape_metric.process_data(data)
ape_statistics = ape_metric.get_all_statistics()
tra_dif = (tra - loc_tra)
# print(tra)
abs_tra_dif = abs((tra - loc_tra)[0]) + abs((tra - loc_tra)[1])
translation = abs(tra[0]) + abs(tra[1])
rot_dif = (rot - loc_rot)
abs_rot_dif = 0
for i in range(0, len(rot_dif)): abs_rot_dif += abs(rot_dif[i][0])+ abs(rot_dif[i][1]) +\
abs(rot_dif[i][2])
# print(abs_tra_dif,abs_rot_dif)
mismatch = abs_tra_dif + abs_rot_dif
tuple = [traj_est, mismatch, traj_est.get_infos()['t_start (s)']]
matches.append(tuple)
matches.sort(key=lambda x: x[2])
segments = [] #The parts of the trajectory are added to this list
for m in matches:
# print(m[1])
if m[1] < 0.1: # if the mismatch is smaller than 1
# print(m[0].get_statistics()['v_avg (m/s)'])
segments.append(m[0])
# print(m[0].get_infos()['t_start (s)'],m[0].get_infos()["path length (m)"])
# print(m[0].get_statistics()['v_avg (m/s)'])
if len(segments) == 0:
print("No matching segments")
whole = trajectory.merge(segments)
traj_ref, traj_est = sync.associate_trajectories(traj,
whole,
max_diff=0.01)
traj_est, rot, tra, _ = trajectory.align_trajectory(traj_est,
traj_ref,
correct_scale=False,
return_parameters=True)
# print(traj_est.get_infos())
return segments, traj_ref, translation