def __init__(self, args):
# parse mocap and save gt in desired format
mocapPath = os.path.join(args.root,
'dataset-{}_512_16'.format(args.seq), 'mav0',
'mocap0', 'data.csv')
groundtruthPath = get_xivo_gt_filename(args.dump, "tumvi", args.seq)
self.saveMocapAs(mocapPath, groundtruthPath)
# Get neural network name, input mode, and covariance prediction type
with open(neural_net + ".pkl", "rb") as fid:
net_params = pickle.load(fid)
network_input_mode = net_params["input_mode"]
network_cov_type = net_params["cov_type"]
assert (network_cov_type == args.cov_type)
neural_net_name = neural_net.split('/')[-1]
# Collect tumvi data
for i, cam_id in enumerate(cam_ids):
for j, seq in enumerate(sequences):
idx_1 = i * len(sequences) + j
estimator_data = get_xivo_output_filename(
covdumpdir, "tumvi", seq, cam_id)
gt_data = get_xivo_gt_filename(covdumpdir, "tumvi", seq)
calib = CovarianceCalibration(
seq,
gt_data,
estimator_data,
three_sigma=False,
start_ind=0,
end_ind=None,
point_cloud_registration='horn',
plot_timesteps=False,
cov_type=network_cov_type,
adjust_startend_to_samplecov=True)
calib.align_gt_to_est()
calib.compute_errors()
num_timesteps = 0
num_1sigma = {cov_type: np.zeros(df[k]) for k, cov_type in enumerate(cov_list)}
num_2sigma = {cov_type: np.zeros(df[k]) for k, cov_type in enumerate(cov_list)}
num_3sigma = {cov_type: np.zeros(df[k]) for k, cov_type in enumerate(cov_list)}
for cam_id in [0, 1]:
for seq in ["room6"]:
print("Cam {}, {}".format(cam_id, seq))
estimator_data = get_xivo_output_filename(args.dump,
"tumvi",
seq,
cam_id=cam_id)
gt_data = get_xivo_gt_filename(args.dump, "tumvi", seq)
if args.cov_source == "original":
adjust = False
else:
adjust = True
calib = CovarianceCalibration(seq,
gt_data,
estimator_data,
three_sigma=False,
start_ind=0,
end_ind=None,
cov_type=args.cov_type,
point_cloud_registration="horn",
adjust_startend_to_samplecov=adjust)
calib.align_gt_to_est()
plot_3D_error_cloud(vel_error, "Velocity Error Cloud (m/s)")
def plot_gauge_group(self):
time_plot(self.time_axis_orig, self.est.gauge_group,
title="Reference Group ID", xlabel=self.time_axis_label)
if __name__=="__main__":
args = parser.parse_args()
# get data files
estimator_data = get_xivo_output_filename(args.dump, args.dataset, args.seq,
cam_id=args.cam_id, sen=args.sen)
gt_data = get_xivo_gt_filename(args.dump, args.dataset, args.seq,
sen=args.sen)
if args.point_cloud_registration=="teaser":
import teaserpp_python as tpp
# avoid analysis for covariance matrices for which we haven't computed a
# sample covariance
if (args.cov_source != "original") and (args.start_ind==0) \
and (args.end_ind is None):
adjust_startend_to_samplecov = True
else:
adjust_startend_to_samplecov = False
# cov_type can only be "all" for original or sampled
if args.cov_type == "all":