Complextext2animation
Complextext2animation copied to clipboard
anindita127
list index out of range
Hello, i just cloned the repo to use it for the pre processing of the KIT dataset but i faced this issue that i didn't know how to solve, if any one knows how please help ?
Traceback (most recent call last):
File "C:\Users\TRETEC\Downloads\Internship documents\Complextext2animation\src\data.py", line 622, in
and the code : class KITMocap(RawData): def init(self, path2data, preProcess_flag=False): super(KITMocap, self).init() # load skeleton
base_path = ''
# Uncomment this when running in DGX everytime
self._SKELPATH = base_path + 'dataProcessing/skeleton.p'
self._MMMSKELPATH = base_path + 'skeleton.xml'
self._MMMSAMPLEPATH = base_path + 'dataProcessing/00001_mmm.xml'
os.makedirs(Path(self._SKELPATH).parent, exist_ok=True)
# get the skeleton and permutation
self.skel, self.permutation, self.new_joints = self.get_skeletonNpermutation()
# save skeleton
torch.save(self.skel, open(self._SKELPATH, 'wb'))
if preProcess_flag:
self.preProcess(path2data)
self.preProcess_axs(path2data)
# Reading data
data = []
# for tup in os.walk(path2data):
tup = [f for f in os.walk(path2data)]
for filename in tup[0][2]:(----------LINE 317 -------)
if Path(filename).suffix == '.xml':
annotpath = Path(
tup[0][0])/(filename.split('_')[0] + '_annotations.json')
annot = json.load(open(annotpath, 'r'))
meta_path = Path(tup[0][0])/(filename.split('_')[0] + '_meta.json')
meta_annot = json.load(open(meta_path, 'r'))
quatpath = filename.split('_')[0] + '_quat.csv'
axpath = filename.split('_')[0] + '_ax.csv'
fkepath = filename.split('_')[0] + '_rifke.fke'
rifkepath = filename.split('_')[0] + '_rifke.csv'
if annot:
kount = 0
for description in annot:
data.append([(Path(tup[0][0])/filename).as_posix(),
description,
(Path(tup[0][0])/quatpath).as_posix(),
(Path(tup[0][0])/axpath).as_posix(),
(Path(tup[0][0])/fkepath).as_posix(),
(Path(tup[0][0])/rifkepath).as_posix(),
meta_annot['annotation_perplexities'][kount]
])
kount += 1
else:
data.append([(Path(tup[0][0])/filename).as_posix(),
'',
(Path(tup[0][0])/quatpath).as_posix(),
(Path(tup[0][0])/axpath).as_posix(),
(Path(tup[0][0])/fkepath).as_posix(),
(Path(tup[0][0])/rifkepath).as_posix(),
''
])
self.df = pd.DataFrame(data=data, columns=[
'euler', 'descriptions', 'quaternion', 'axis-angle', 'fke', 'rifke', 'perplexity'])
self.columns = pd.read_csv(
self.df.iloc[0].quaternion, index_col=0).columns
joints = [col[:-3] for col in self.columns]
self.joints = []
self.columns_dict = {}
start = 0
for joint in joints:
if not self.joints:
self.joints.append(joint)
end = 1
elif self.joints[-1] == joint:
end += 1
else:
self.columns_dict.update(
{self.joints[-1]: self.columns[start:end]})
self.joints.append(joint)
start = end
end = end + 1
self.columns_dict.update({self.joints[-1]: self.columns[start:end]})
def _get_df(self):
return self.df
def _get_f(self):
return 100
def _get_skeleton(self):
return self.skel
@property
def rifke_dict(self):
return {'fid_l': np.array([14, 15]),
'fid_r': np.array([19, 20]),
'sdr_l': 6,
'sdr_r': 9,
'hip_l': 12,
'hip_r': 17}
def preProcess(self, path2data):
print('Preprocessing KIT Data')
for tup in os.walk(path2data):
for filename in (tup[2]):
if Path(filename).suffix == '.xml':
filepath = Path(tup[0])/filename
quatpath = filename.split('_')[0] + '_quat.csv'
quatpath = (Path(tup[0])/quatpath).as_posix()
xyz_data, skel, joints, root_pos, rotations = self.mmm2quat(
filepath)
# create quat dataframe
root_pos = root_pos.squeeze(0)
rotations = rotations.contiguous().view(
rotations.shape[1], -1)
quats = torch.cat([root_pos, rotations], dim=-1).numpy()
columns = ['root_tx', 'root_ty', 'root_tz'] + \
['{}_{}'.format(joint, axis) for joint in joints for axis in [
'rw', 'rx', 'ry', 'rz']]
df = pd.DataFrame(data=quats, columns=columns)
df.to_csv(quatpath)
filename_fke = str(Path(quatpath).parent) + \
'/' + filename.split('_')[0] + '_fke.csv'
print(filename_fke)
filename_rifke = str(
Path(quatpath).parent) + '/' + filename.split('_')[0] + '_rifke.csv'
self.quat2fke(df, filename_fke, filename_rifke)
def preProcess_axs(self, path2data):
print('Preprocessing KIT Data axis angles')
for tup in os.walk(path2data):
for filename in (tup[2]):
if Path(filename).suffix == '.xml':
filepath = Path(tup[0])/filename
axpath = filename.split('_')[0] + '_ax.csv'
axpath = (Path(tup[0])/axpath).as_posix()
# xyz_data, skel, joints, root_pos, rotations = self.mmm2quat(
# filepath)
joints, root_pos, rotations = self.mmm2axis(filepath)
# create quat dataframe
root_pos = root_pos.squeeze(0)
rotations = rotations.contiguous().view(
rotations.shape[1], -1)
quats = torch.cat([root_pos, rotations], dim=-1).numpy()
columns = ['root_tx', 'root_ty', 'root_tz'] + \
['{}_{}'.format(joint, axis) for joint in joints for axis in [
'rx', 'ry', 'rz']]
df = pd.DataFrame(data=quats, columns=columns)
df.to_csv(axpath)
def mat2amc(self, data, filename):
lines = ["#!OML:ASF H:",
":FULLY-SPECIFIED",
":DEGREES"]
for count, row in enumerate(data):
start = 0
lines.append('{}'.format(count+1))
for joint in self.joints:
end = start + len(self.columns_dict[joint])
format_str = '{} ' * (len(self.columns_dict[joint]) + 1)
format_str = format_str[:-1] # remove the extra space
lines.append(format_str.format(
*([joint] + list(row[start:end]))))
start = end
lines = '\n'.join(lines) + '\n'
os.makedirs(filename.parent, exist_ok=True)
with open(filename, 'w') as fp:
fp.writelines(lines)
def get_new_parents(self, parents, joints_left, joints_right, joints):
permutation, new_parents = permute(parents)
joints_w_root = ['root'] + joints
new_joints = [joints_w_root[perm] for perm in permutation]
new_joints_idx = list(range(len(new_joints)))
new_joints_left = []
new_joints_right = []
for idx, jnt in enumerate(new_joints):
if jnt[0] == 'R':
new_joints_right.append(idx)
else:
new_joints_left.append(idx)
return permutation, new_parents, new_joints_left, new_joints_right, new_joints
# KITMocap Specific
def get_skeletonNpermutation(self):
# make a parents_list
parents = [-1, 3, 0, 2, 1, 8, 9, 0, 7, 1,
6, 12, 5, 16, 17, 0, 15, 1, 14, 20, 13]
joints_left = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12]
joints_right = [13, 14, 15, 16, 17, 18, 19, 20]
# read a demo file to get the joints
joints, _, _, _ = self.mmm2csv(Path(self._MMMSAMPLEPATH))
permutation, new_parents, new_joints_left, new_joints_right, new_joints = self.get_new_parents(
parents, joints_left, joints_right, joints)
import xml.etree.ElementTree as ET
tree = ET.parse(self._MMMSKELPATH)
root = tree.getroot()
# make an offset list
offset_dict = self.get_offsets(root, joints)
offset_dict.update({'root': [0, 0, 0]})
new_offsets = [offset_dict[joint] for joint in new_joints]
# make a Skeleton
skel = Skeleton(new_offsets, new_parents,
new_joints_left, new_joints_right, new_joints)
return skel, permutation, new_joints
# read an xml file
def mmm2csv(self, src):
joint_names, mmm_dict = parse_motions(src.as_posix())[0]
root_pos = np.array(mmm_dict['root_pos'],
dtype=float) # * 0.001 / 0.056444
#root_pos = root_pos[:, [1,2,0]]
root_rot = np.array(mmm_dict['root_rot'],
dtype=float) # * 180/np.pi
#root_rot = root_rot[:, [1,2,0]]
joint_pos = np.array(
mmm_dict['joint_pos'], dtype=float) # * 180/np.pi
joint_dict = {}
for idx, name in enumerate(joint_names):
if name.split('_')[0][-1] != 't':
xyz = name.split('_')[0][-1]
joint = name.split('_')[0][:-1]
else:
xyz = 'y'
joint = name.split('_')[0]
if joint not in joint_dict:
joint_dict[joint] = dict()
joint_dict[joint][xyz] = joint_pos[:, idx]
joints = []
values = []
for cnt, joint in enumerate(joint_dict):
joint_vals = []
joints.append(joint)
for axes in ['x', 'y', 'z']:
if axes in joint_dict[joint]:
joint_vals.append(joint_dict[joint][axes])
else:
joint_vals.append(np.zeros_like(root_pos[:, 0]))
values.append(np.stack(joint_vals, axis=1))
values = np.stack(values, axis=0)
return joints, root_pos, root_rot, values
def get_offsets(self, root, Joints):
joints = root.findall('RobotNode')
offset_dict = {}
for joint in joints:
matrix = joint.findall('Transform')
if matrix:
offset = []
# switch y and z axis
for row in ['row1', 'row3', 'row2']:
Row = matrix[0].findall('Matrix4x4')[0].findall(row)
offset.append(float(Row[0].attrib['c4']))
joint_name = joint.attrib['name']
if joint_name.split('_')[0][-6:] == 'egment':
if joint_name[:-13] in Joints:
offset_dict[joint_name[:-13]] = offset
else:
if joint_name[:-6] in Joints:
offset_dict[joint_name[:-6]] = offset
elif joint_name[:-7] in Joints:
offset_dict[joint_name[:-7]] = offset
return offset_dict
def mmm2axis(self, path):
joints, root_pos, root_rot, values = self.mmm2csv(path)
# convert to quaternions
values_quat = euler2quatbatch(values, axes='sxyz')
root_rot_quat = euler2quatbatch(root_rot, axes='sxyz')
# switch y and z axis
# Note the qinv_np is very important as 2 axes are being interchanged - can be proved using basic vector equations
root_pos = root_pos[..., [0, 2, 1]]
values_quat = qinv_np(values_quat[..., [0, 1, 3, 2]])
root_rot_quat = qinv_np(root_rot_quat[..., [0, 1, 3, 2]])
rotations = np.expand_dims(np.transpose(np.concatenate((np.expand_dims(
root_rot_quat, axis=0), values_quat), axis=0), axes=[1, 0, 2]), axis=0)
root_pos = np.expand_dims(root_pos, axis=0)
axis_rotations = quaternion_to_expmap(rotations)
new_rotations = torch.from_numpy(
axis_rotations[:, :, self.permutation, :])
new_root_pos = torch.from_numpy(root_pos.copy())
return self.new_joints, new_root_pos, new_rotations
def mmm2quat(self, path):
joints, root_pos, root_rot, values = self.mmm2csv(path)
# convert to quaternions
values_quat = euler2quatbatch(values, axes='sxyz')
root_rot_quat = euler2quatbatch(root_rot, axes='sxyz')
# switch y and z axis
# Note the qinv_np is very important as 2 axes are being interchanged - can be proved using basic vector equations
root_pos = root_pos[..., [0, 2, 1]]
values_quat = qinv_np(values_quat[..., [0, 1, 3, 2]])
root_rot_quat = qinv_np(root_rot_quat[..., [0, 1, 3, 2]])
rotations = np.expand_dims(np.transpose(np.concatenate((np.expand_dims(
root_rot_quat, axis=0), values_quat), axis=0), axes=[1, 0, 2]), axis=0)
root_pos = np.expand_dims(root_pos, axis=0)
new_rotations = torch.from_numpy(rotations[:, :, self.permutation, :])
new_root_pos = torch.from_numpy(root_pos.copy())
xyz_data = self.skel.forward_kinematics(new_rotations, new_root_pos)[0]
return xyz_data.numpy(), self.skel, self.new_joints, new_root_pos, new_rotations
if name == 'main': """PreProcessing""" import argparse parser = argparse.ArgumentParser() parser.add_argument('-dataset', default='KITMocap', type=str, help='dataset kind') parser.add_argument('-path2data', default='./dataset/kit-mocap', type=str, help='dataset kind') args, _ = parser.parse_known_args() eval(args.dataset)(args.path2data, preProcess_flag=True)(-----------------LINE622----------------) print('Succesfully Preprocessed {} data'.format(args.dataset))
Hi. This error usually occurs when it cannot find the data. Please make sure you have copied the dataset to the correct location.
Thank u for ur fast reply, but i did exactly like this " Download the KIT Motion-Language dataset and unzip the files in the $BASE/dataset/kit-mocap folder." and my repository looks like this Complextext2animation dataset kit-mocap src src dataProcessing
am i doing somthing wrong ?
Complextext2animation ----------dataset --------------kit-mocap ----------src --------------src -------------------dataProcessing
There should be one src folder. Complextext2animation ----------dataset --------------kit-mocap ----------src --------------dataProcessing
Thank u for ur fast reply, but i did exactly like this " Download the KIT Motion-Language dataset and unzip the files in the $BASE/dataset/kit-mocap folder." and my repository looks like this Complextext2animation dataset kit-mocap src src dataProcessing
am i doing somthing wrong ?
I encounter the same problem with you. i guess the reason is the "import package" of python is messed up, the program cannot find the existed file. did you fix the problem? i try many different ways, but all of them do not work!
i figure out. at the root dir (which is $bas3/Complextext2motion), run "python src/dataProcessing/meanVariance.py -mask '[0]' -feats_kind rifke -dataset KITMocap -path2data dataset/kit-mocap -f_new 8"
just give it a try