def __init__(self, batch_sz, betas=None, pose=None, trans=None, offsets=None, faces=None, gender='male'):
super(th_batch_SMPL, self).__init__()
if betas is None:
self.betas = nn.Parameter(torch.zeros(batch_sz, 300))
else:
assert betas.ndim == 2
self.betas = nn.Parameter(betas)
if pose is None:
self.pose = nn.Parameter(torch.zeros(batch_sz, 72))
else:
assert pose.ndim == 2
self.pose = nn.Parameter(pose)
if trans is None:
self.trans = nn.Parameter(torch.zeros(batch_sz, 3))
else:
assert trans.ndim == 2
self.trans = nn.Parameter(trans)
if offsets is None:
self.offsets = nn.Parameter(torch.zeros(batch_sz, 6890,3))
else:
assert offsets.ndim == 3
self.offsets = nn.Parameter(offsets)
self.faces = faces
self.gender = gender
# pytorch smpl
self.smpl = SMPL_Layer(center_idx=0, gender=gender,
model_root='/BS/bharat/work/installation/smplpytorch/smplpytorch/native/models')
# Landmarks
self.body25_reg_torch, self.face_reg_torch, self.hand_reg_torch = torch_pose_obj_data(batch_size=batch_sz)
def __init__(self,
betas=None,
pose=None,
trans=None,
offsets=None,
gender='male'):
super(th_SMPL, self).__init__()
if betas is None:
self.betas = nn.Parameter(torch.zeros(300, ))
else:
self.betas = nn.Parameter(betas)
if pose is None:
self.pose = nn.Parameter(torch.zeros(72, ))
else:
self.pose = nn.Parameter(pose)
if trans is None:
self.trans = nn.Parameter(torch.zeros(3, ))
else:
self.trans = nn.Parameter(trans)
if offsets is None:
self.offsets = nn.Parameter(torch.zeros(6890, 3))
else:
self.offsets = nn.Parameter(offsets)
## pytorch smpl
self.smpl = SMPL_Layer(
center_idx=0,
gender=gender,
model_root=
'/BS/bharat/work/installation/smplpytorch/smplpytorch/native/models'
)
def __init__(self,
batch_sz,
top_betas=None,
other_betas=None,
global_pose=None,
other_pose=None,
trans=None,
offsets=None,
faces=None,
gender='male'):
super(th_batch_SMPL_split_params, self).__init__()
if top_betas is None:
self.top_betas = nn.Parameter(torch.zeros(batch_sz, 2))
else:
assert top_betas.ndim == 2
self.top_betas = nn.Parameter(top_betas)
if other_betas is None:
self.other_betas = nn.Parameter(torch.zeros(batch_sz, 298))
else:
assert other_betas.ndim == 2
self.other_betas = nn.Parameter(other_betas)
if global_pose is None:
self.global_pose = nn.Parameter(torch.zeros(batch_sz, 3))
else:
assert global_pose.ndim == 2
self.global_pose = nn.Parameter(global_pose)
if other_pose is None:
self.other_pose = nn.Parameter(torch.zeros(batch_sz, 69))
else:
assert other_pose.ndim == 2
self.other_pose = nn.Parameter(other_pose)
if trans is None:
self.trans = nn.Parameter(torch.zeros(batch_sz, 3))
else:
assert trans.ndim == 2
self.trans = nn.Parameter(trans)
if offsets is None:
self.offsets = nn.Parameter(torch.zeros(batch_sz, 6890, 3))
else:
assert offsets.ndim == 3
self.offsets = nn.Parameter(offsets)
self.betas = torch.cat([self.top_betas, self.other_betas], axis=1)
self.pose = torch.cat([self.global_pose, self.other_pose], axis=1)
self.faces = faces
self.gender = gender
# pytorch smpl
self.smpl = SMPL_Layer(center_idx=0,
gender=gender,
model_root='./body_models/smpl')
# Landmarks
self.body25_reg_torch, self.face_reg_torch, self.hand_reg_torch = torch_pose_obj_data(
batch_size=batch_sz)
def __init__(self, folder, device, gender='male'):
super(VolumetricSMPL, self).__init__()
with torch.no_grad():
# Load transformation
with open(join(folder, 'scale_center.pkl'), 'rb') as f:
self.scale, self.center = pkl.load(f, encoding='latin-1')
self.scale = torch.tensor(self.scale.astype('float32'),
requires_grad=False).to(device)
self.center = torch.tensor(self.center.astype('float32'),
requires_grad=False).to(device)
# Load closest_point
with open(join(folder, 'closest_point.pkl'), 'rb') as f:
# Shape: 3(x, y, z) x res x res x res
closest_point = pkl.load(f,
encoding='latin-1').astype('float32')
self.closest_point = torch.tensor(closest_point,
requires_grad=False).permute(
3, 0, 1, 2).to(device)
res = self.closest_point.shape[-1]
# Load shapedirs
with open(join(folder, 'shapedirs.pkl'), 'rb') as f:
# Shape: 10 x 3(x, y, z) x res x res x res
shapedirs = pkl.load(f, encoding='latin-1')[..., :10].astype(
'float32') # keep only dim shape dims
self.shapedirs = torch.tensor(
shapedirs.reshape(res, res, res, -1),
requires_grad=False).permute(3, 0, 1, 2).to(device)
# Load posedirs
with open(join(folder, 'posedirs.pkl'), 'rb') as f:
# Shape: 207 x 3 x res x res x res
posedirs = pkl.load(f, encoding='latin-1').astype('float32')
self.posedirs = torch.tensor(
posedirs.reshape(res, res, res, -1),
requires_grad=False).permute(3, 0, 1, 2).to(device)
# Load skinning_weights
with open(join(folder, 'skinning_weights.pkl'), 'rb') as f:
skinning_weights = pkl.load(
f, encoding='latin-1').astype('float32')
self.skinning_weights = torch.tensor(
skinning_weights,
requires_grad=False).permute(3, 0, 1, 2).to(device)
self.grid_fn = correspondence_to_smpl_function
## pytorch smpl
self.smpl = SMPL_Layer(
center_idx=0,
gender=gender,
model_root='smplpytorch/smplpytorch/native/models').to(device)