def main(args):
if args.verbose:
logging.basicConfig(level=logging.DEBUG)
args.disable_cache = True # we do not want to load stale data here!
path = args.dataset
if not os.path.exists(path):
exit('dataset at path "%s" does not exist' % path)
# check and export-all does not need the loaded dataset
if args.action == 'check':
check(args)
return
elif args.action == 'export-all':
export_all(args)
return
print('loading data set "%s" ...' % path)
dataset = data.load_dataset_from_args(args)
action_func = None
if args.action == 'report':
action_func = report
elif args.action == 'export':
action_func = export
elif args.action == 'plot':
action_func = plot
assert action_func is not None
action_func(dataset, args)
def main(args):
if args.verbose:
logging.basicConfig(level=logging.DEBUG)
args.disable_cache = True # we do not want to load stale data here!
path = args.dataset
if not os.path.exists(path):
exit('dataset at path "%s" does not exist' % path)
# check and export-all does not need the loaded dataset
if args.action == 'check':
check(args)
return
elif args.action == 'export-all':
export_all(args)
return
print('loading data set "%s" ...' % path)
dataset = data.load_dataset_from_args(args)
action_func = None
if args.action == 'report':
action_func = report
elif args.action == 'export':
action_func = export
elif args.action == 'plot':
action_func = plot
assert action_func is not None
action_func(dataset, args)
def merge(args):
# Merges two datasets into one
dataset1 = data.load_dataset_from_args(args.dataset)
dataset2 = data.load_dataset_from_args(args.other_dataset)
dataset1.merge_with_dataset(dataset2)
def export_all(args):
# Load MMM dataset WITHOUT normalization
args.motion_type = 'mmm-nlopt'
args.disable_normalization = True
args.disable_smoothing = True
print('Loading MMM data without normalization ...')
mmm = data.load_dataset_from_args(args)
print('done, %d motions and %d features loaded' %
(mmm.n_samples, mmm.n_features))
print('')
# Load MMM dataset WITH normalization
args.motion_type = 'mmm-nlopt'
args.disable_normalization = False
args.disable_smoothing = True
print('Loading MMM data with normalization ...')
normalized_mmm = data.load_dataset_from_args(args)
normalized_mmm.feature_names = [
'normalized_' + name for name in normalized_mmm.feature_names
]
print('done, %d motions and %d features loaded' %
(normalized_mmm.n_samples, normalized_mmm.n_features))
print('')
# Load Vicon dataset
args.motion_type = 'vicon'
args.disable_normalization = True
args.disable_smoothing = True
print('Loading vicon data without normalization ...')
vicon = data.load_dataset_from_args(args)
print('done, %d motions and %d features loaded' %
(vicon.n_samples, vicon.n_features))
print('')
# Manually perform normalization on Vicon using the root_rot and root_pos information
print('Performing normalization on vicon dataset ...')
normalized_vicon = vicon.copy()
root_pos_indexes = mmm.indexes_for_feature('root_pos')
root_rot_indexes = mmm.indexes_for_feature('root_rot')
for sample_idx in xrange(normalized_vicon.n_samples):
X_curr = normalized_vicon.X[sample_idx]
n_frames = X_curr.shape[0]
marker_pos_indexes = normalized_vicon.indexes_for_feature('marker_pos')
assert len(marker_pos_indexes) % 3 == 0
n_markers = len(marker_pos_indexes) / 3
root_pos0 = mmm.X[sample_idx][
0, root_pos_indexes] # do not use the normalized dataset here!
root_rot = mmm.X[
sample_idx][:,
root_rot_indexes] # do not use the normalized dataset here!
assert root_rot.shape == (n_frames, 3)
# Normalize marker positions
root_pose0_inv = np.linalg.inv(
pose_matrix(root_pos0, (0., 0., root_rot[0][2])))
for marker_idx in xrange(n_markers):
start_idx = marker_pos_indexes[0] + marker_idx * 3
end_idx = start_idx + 3
marker_pos = X_curr[:, start_idx:end_idx]
n_samples = marker_pos.shape[0]
marker_pos_plus_one = np.hstack((marker_pos, np.ones(
(n_samples, 1))))
assert marker_pos_plus_one.shape == (n_samples, 4)
normalized_marker_pos = np.dot(root_pose0_inv,
marker_pos_plus_one.T).T[:, 0:3]
assert normalized_marker_pos.shape == marker_pos.shape
X_curr[:, start_idx:end_idx] = normalized_marker_pos
# Normalize velocities and accelerations
marker_vel_indexes = normalized_vicon.indexes_for_feature('marker_vel')
marker_acc_indexes = normalized_vicon.indexes_for_feature('marker_acc')
for idx in xrange(n_frames):
root_rot_inv = np.linalg.inv(
rotation_matrix(root_rot[idx][0], root_rot[idx][1],
root_rot[idx][2]))
for marker_idx in xrange(n_markers):
vel_start_idx = marker_vel_indexes[0] + marker_idx * 3
vel_end_idx = vel_start_idx + 3
vel = X_curr[idx, vel_start_idx:vel_end_idx]
X_curr[idx,
vel_start_idx:vel_end_idx] = np.dot(root_rot_inv, vel)
acc_start_idx = marker_acc_indexes[0] + marker_idx * 3
acc_end_idx = acc_start_idx + 3
acc = X_curr[idx, acc_start_idx:acc_end_idx]
X_curr[idx,
acc_start_idx:acc_end_idx] = np.dot(root_rot_inv, acc)
# No need to normalize marker_vel_norm and marker_acc_norm
normalized_vicon.feature_names = [
'normalized_' + name for name in normalized_vicon.feature_names
]
print('done, %d motions and %d features loaded' %
(normalized_vicon.n_samples, normalized_vicon.n_features))
print('')
# Merge all datasets into one
print('merging datasets ...')
final_dataset = mmm.copy()
final_dataset.merge_with_dataset(normalized_mmm)
final_dataset.merge_with_dataset(vicon)
final_dataset.merge_with_dataset(normalized_vicon)
assert final_dataset.feature_names == mmm.feature_names + normalized_mmm.feature_names + vicon.feature_names + normalized_vicon.feature_names
assert final_dataset.n_features == mmm.n_features + normalized_mmm.n_features + vicon.n_features + normalized_vicon.n_features
print('done, %d motions and %d features' %
(final_dataset.n_samples, final_dataset.n_features))
print('')
if args.output is not None:
print('saving dataset ...')
with open(args.output, 'wb') as f:
pickle.dump(final_dataset, f)
print('done')
def merge(args):
# Merges two datasets into one
dataset1 = data.load_dataset_from_args(args.dataset)
dataset2 = data.load_dataset_from_args(args.other_dataset)
dataset1.merge_with_dataset(dataset2)
def export_all(args):
# Load MMM dataset WITHOUT normalization
args.motion_type = 'mmm-nlopt'
args.disable_normalization = True
args.disable_smoothing = True
print('Loading MMM data without normalization ...')
mmm = data.load_dataset_from_args(args)
print('done, %d motions and %d features loaded' % (mmm.n_samples, mmm.n_features))
print('')
# Load MMM dataset WITH normalization
args.motion_type = 'mmm-nlopt'
args.disable_normalization = False
args.disable_smoothing = True
print('Loading MMM data with normalization ...')
normalized_mmm = data.load_dataset_from_args(args)
normalized_mmm.feature_names = ['normalized_' + name for name in normalized_mmm.feature_names]
print('done, %d motions and %d features loaded' % (normalized_mmm.n_samples, normalized_mmm.n_features))
print('')
# Load Vicon dataset
args.motion_type = 'vicon'
args.disable_normalization = True
args.disable_smoothing = True
print('Loading vicon data without normalization ...')
vicon = data.load_dataset_from_args(args)
print('done, %d motions and %d features loaded' % (vicon.n_samples, vicon.n_features))
print('')
# Manually perform normalization on Vicon using the root_rot and root_pos information
print('Performing normalization on vicon dataset ...')
normalized_vicon = vicon.copy()
root_pos_indexes = mmm.indexes_for_feature('root_pos')
root_rot_indexes = mmm.indexes_for_feature('root_rot')
for sample_idx in xrange(normalized_vicon.n_samples):
X_curr = normalized_vicon.X[sample_idx]
n_frames = X_curr.shape[0]
marker_pos_indexes = normalized_vicon.indexes_for_feature('marker_pos')
assert len(marker_pos_indexes) % 3 == 0
n_markers = len(marker_pos_indexes) / 3
root_pos0 = mmm.X[sample_idx][0, root_pos_indexes] # do not use the normalized dataset here!
root_rot = mmm.X[sample_idx][:, root_rot_indexes] # do not use the normalized dataset here!
assert root_rot.shape == (n_frames, 3)
# Normalize marker positions
root_pose0_inv = np.linalg.inv(pose_matrix(root_pos0, (0., 0., root_rot[0][2])))
for marker_idx in xrange(n_markers):
start_idx = marker_pos_indexes[0] + marker_idx * 3
end_idx = start_idx + 3
marker_pos = X_curr[:, start_idx:end_idx]
n_samples = marker_pos.shape[0]
marker_pos_plus_one = np.hstack((marker_pos, np.ones((n_samples, 1))))
assert marker_pos_plus_one.shape == (n_samples, 4)
normalized_marker_pos = np.dot(root_pose0_inv, marker_pos_plus_one.T).T[:, 0:3]
assert normalized_marker_pos.shape == marker_pos.shape
X_curr[:, start_idx:end_idx] = normalized_marker_pos
# Normalize velocities and accelerations
marker_vel_indexes = normalized_vicon.indexes_for_feature('marker_vel')
marker_acc_indexes = normalized_vicon.indexes_for_feature('marker_acc')
for idx in xrange(n_frames):
root_rot_inv = np.linalg.inv(rotation_matrix(root_rot[idx][0], root_rot[idx][1], root_rot[idx][2]))
for marker_idx in xrange(n_markers):
vel_start_idx = marker_vel_indexes[0] + marker_idx * 3
vel_end_idx = vel_start_idx + 3
vel = X_curr[idx, vel_start_idx:vel_end_idx]
X_curr[idx, vel_start_idx:vel_end_idx] = np.dot(root_rot_inv, vel)
acc_start_idx = marker_acc_indexes[0] + marker_idx * 3
acc_end_idx = acc_start_idx + 3
acc = X_curr[idx, acc_start_idx:acc_end_idx]
X_curr[idx, acc_start_idx:acc_end_idx] = np.dot(root_rot_inv, acc)
# No need to normalize marker_vel_norm and marker_acc_norm
normalized_vicon.feature_names = ['normalized_' + name for name in normalized_vicon.feature_names]
print('done, %d motions and %d features loaded' % (normalized_vicon.n_samples, normalized_vicon.n_features))
print('')
# Merge all datasets into one
print('merging datasets ...')
final_dataset = mmm.copy()
final_dataset.merge_with_dataset(normalized_mmm)
final_dataset.merge_with_dataset(vicon)
final_dataset.merge_with_dataset(normalized_vicon)
assert final_dataset.feature_names == mmm.feature_names + normalized_mmm.feature_names + vicon.feature_names + normalized_vicon.feature_names
assert final_dataset.n_features == mmm.n_features + normalized_mmm.n_features + vicon.n_features + normalized_vicon.n_features
print('done, %d motions and %d features' % (final_dataset.n_samples, final_dataset.n_features))
print('')
if args.output is not None:
print('saving dataset ...')
with open(args.output, 'wb') as f:
pickle.dump(final_dataset, f)
print('done')
