def _build_training_set_index(self, radius=2):
self.training_set_index = []
self.poses = {}
self.calib = {}
for drive in self.sequences:
trajectory = self._read_oxts_data(drive)
proj_c2p, proj_v2c, imu2cam = self._read_raw_calib_data(drive)
for i in range(len(trajectory)):
trajectory[i] = np.dot(imu2cam, util.inv_SE3(trajectory[i]))
trajectory[i][0:3, 3] *= self.args['scale']
self.poses[drive] = trajectory
self.calib[drive] = (proj_c2p, proj_v2c, imu2cam)
for i in range(len(trajectory)):
seq = []
for j in range(i-radius, i+radius+1):
j = min(max(0, j), len(trajectory)-1)
frame = {
'image': self._fetch_image_path(drive, j),
'velo': self._fetch_velo_path(drive, j),
'pose': self.poses[drive][j],
'drive': drive,
}
seq.append(frame)
self.training_set_index.append(seq)
def test_drive(self, drive, radius=2):
self.poses = {}
self.calib = {}
trajectory = self._read_oxts_data(drive)
proj_c2p, proj_v2c, imu2cam = self._read_raw_calib_data(drive)
for i in range(len(trajectory)):
trajectory[i] = np.dot(imu2cam, util.inv_SE3(trajectory[i]))
trajectory[i][0:3, 3] *= self.args['scale']
self.poses[drive] = trajectory
self.calib[drive] = (proj_c2p, proj_v2c, imu2cam)
for i in range(len(trajectory)):
seq = []
for j in range(i-radius, i+radius+1):
j = min(max(0, j), len(trajectory)-1)
frame = {
'image': self._fetch_image_path(drive, j),
'velo': self._fetch_velo_path(drive, j),
'pose': self.poses[drive][j],
'drive': drive,
}
seq.append(frame)
yield self._load_example(seq)
def test_set_iterator(self, radius=2):
test_list = []
with open('data/kitti/test_files_eigen.txt') as f:
reader = csv.reader(f)
for row in reader:
test_list.append(row[0])
self.poses = {}
self.calib = {}
for test_frame in test_list:
comps = test_frame.split('/')
drive = comps[1].replace('_sync', '')
frame = int(comps[4].replace('.png', ''))
if drive not in self.poses:
trajectory = self._read_oxts_data(drive)
proj_c2p, proj_v2c, imu2cam = self._read_raw_calib_data(drive)
for i in range(len(trajectory)):
trajectory[i] = np.dot(imu2cam, util.inv_SE3(trajectory[i]))
trajectory[i][0:3, 3] *= self.args['scale']
self.poses[drive] = trajectory
self.calib[drive] = (proj_c2p, proj_v2c, imu2cam)
else:
trajectory = self.poses[drive]
seq = []
for j in range(frame-radius, frame+radius+1):
j = min(max(0, j), len(trajectory)-1)
frame = {
'image': self._fetch_image_path(drive, j),
'velo': self._fetch_velo_path(drive, j),
'pose': self.poses[drive][j],
'drive': drive,
}
seq.append(frame)
data_blob = self._load_example(seq)
yield data_blob
def _build_sequence_set_index(self, tskip=0.6):
self.sequence_set_index = []
for sceneb in self._collect_train_scenes(mode='train'):
seach_str = os.path.join(self.dataset_path, '%s*' % sceneb)
for scene in glob.glob(seach_str):
scene = os.path.basename(scene)
try:
timestamps, pairs, poses = self._load_trajectory_data(
scene)
except:
continue
tstamps = np.array(timestamps, dtype="float64")
tcur = tstamps[0] - 2 * tskip
for i in range(0, len(timestamps)):
if tstamps[i] - tcur > tskip:
tcur = tstamps[i]
vid = self._make_video(i,
tstamps,
radius=self.args.get('radius'))
sequence = []
for j in range(len(vid)):
t = timestamps[vid[j]]
frame = {
'image':
os.path.join(self.dataset_path, scene,
pairs[t][0]),
'depth':
os.path.join(self.dataset_path, scene,
pairs[t][1]),
'pose':
util.inv_SE3(poses[t]),
}
sequence.append(frame)
self.sequence_set_index.append(sequence)