def __getitem__(self, index):
data_numpy = self.data[index]
label = self.label[index]
data_numpy = np.array(data_numpy)
C, T, V, M = data_numpy.shape
if self.normalization:
data_numpy = (data_numpy - self.mean_map) / self.std_map
if self.random_shift:
data_numpy = tools.random_shift(data_numpy)
if self.random_choose:
data_numpy = tools.random_choose(data_numpy, self.window_size)
elif self.window_size > 0:
data_numpy = tools.auto_pading(data_numpy, self.window_size)
if self.random_move:
data_numpy = tools.random_move(data_numpy)
if self.argument:
if V == 18:
conf = data_numpy[2].reshape(1, -1, V, M)
data_numpy = data_numpy[:2]
motion = np.concatenate((data_numpy, np.zeros(
(C - 1, 1, V, M))), 1) - np.concatenate((np.zeros(
(C - 1, 1, V, M)), data_numpy), 1)
motion = np.concatenate(
(motion[:, 1:-1, :, :], np.zeros((C - 1, 1, V, M))), 1)
data_numpy = np.concatenate((data_numpy, motion, conf), 0)
else:
motion = np.concatenate((data_numpy, np.zeros(
(C, 1, V, M))), 1) - np.concatenate((np.zeros(
(C, 1, V, M)), data_numpy), 1)
motion = np.concatenate(
(motion[:, 1:-1, :, :], np.zeros((C, 1, V, M))), 1)
data_numpy = np.concatenate((data_numpy, motion), 0)
return data_numpy, label, index
def __getitem__(self, index):
if "train" in self.data_path and "bone" in self.data_path:
train_data_file = train_data_bone_files[index]
filename = train_data_file
data_numpy = np.load(train_data_bone_path + train_data_file)
label = int(
filename[filename.index("A") + 1:filename.index("A") + 4]) - 1
elif "test" in self.data_path and "bone" in self.data_path:
test_data_file = test_data_bone_files[index]
filename = test_data_file
data_numpy = np.load(test_data_bone_path + test_data_file)
label = int(
filename[filename.index("A") + 1:filename.index("A") + 4]) - 1
data_numpy = data_numpy[:, :, :67, :]
if self.normalization:
data_numpy = (data_numpy - self.mean_map) / self.std_map
if self.random_shift:
data_numpy = tools.random_shift(data_numpy)
if self.random_choose:
data_numpy = tools.random_choose(data_numpy, self.window_size)
elif self.window_size > 0:
data_numpy = tools.auto_pading(data_numpy, self.window_size)
if self.random_move:
data_numpy = tools.random_move(data_numpy)
return data_numpy, label, index
def __getitem__(self, index):
data_numpy = self.data[index]
label = int(self.label[index])
data_numpy = np.array(data_numpy)
if self.normalization:
data_numpy = (data_numpy - self.mean_map) / self.std_map
if self.random_shift:
data_numpy = tools.random_shift(data_numpy)
if self.random_choose:
data_numpy = tools.random_choose(data_numpy, self.window_size)
elif self.window_size > 0:
data_numpy = tools.auto_pading(data_numpy, self.window_size)
if self.random_move:
data_numpy = tools.random_move(data_numpy)
# # flipling
# if self.flip:
# data_numpy = tools.flip(data_numpy)
# if self.rescale:
# data_numpy = tools.rescale(data_numpy)
# if self.rotation:
# data_numpy = tools.rotation(data_numpy)
# if self.aug:
# data_numpy = tools.augmentation(data_numpy)
if self.pad:
data_numpy = tools.temporal_pad(data_numpy)
if self.is_test:
return data_numpy, label, index
else:
return {
"batchdata": torch.FloatTensor(data_numpy),
"label": torch.LongTensor([label])
}
def __getitem__(self, index):
"""
self.data is fixed data.
each epoch, the sequence of index shuffled.
ex) epoch0 --> [ 1 3 2 4 0]
ex) epoch1 --> [ 0 3 4 2 1]
:param index:
:return:
"""
_data_numpy = self.data[index] # sample index.
label = self.label[index]
data_numpy = np.array(_data_numpy) # 3, 300, 25, 2
if self.modf == 'knee':
data_numpy = self.tmp_noise_knee_test(data_numpy)
if self.modf == 'shld':
data_numpy = self.tmp_noise_shld_train(data_numpy)
if self.modf == 'noise':
if self.ttype == 'All': # all
data_numpy = self.noise_joint_all(data_numpy)
else: # Fixed
data_numpy = self.noise_joint_fix(data_numpy)
if self.modf == 'mutual':
if self.ttype == 'static':
__data_numpy = self.noise_joint_static(data_numpy,
self.drop_num)
else:
dynamic_drop_num = np.random.randint(1, self.drop_num + 1)
__data_numpy = self.noise_joint_static(data_numpy,
dynamic_drop_num)
data_numpy = np.array([_data_numpy, __data_numpy])
if self.normalization:
data_numpy = (data_numpy - self.mean_map) / self.std_map
if self.random_shift:
data_numpy = tools.random_shift(data_numpy)
if self.random_choose:
data_numpy = tools.random_choose(data_numpy, self.window_size)
elif self.window_size > 0:
data_numpy = tools.auto_pading(data_numpy, self.window_size)
if self.random_move:
data_numpy = tools.random_move(data_numpy)
# print(f'getitem {index} after', data_numpy[:,0,14,0 ])
return data_numpy, label, index
def __getitem__(self, index):
# data_numpy = self.data[index]
# label = self.label[index]
# data_numpy = np.array(data_numpy)
'''
if "train" in self.data_path and "joint_motion" in self.data_path:
train_data_file = train_data_joint_motion_files[index]
filename = train_data_file
f = h5py.File(train_data_joint_motion_path + train_data_file, "r")
data_numpy = f[list(f.keys())[-1]][:]
label =int(filename[filename.index("A") + 1 : filename.index("A") + 4]) - 1
f.close()
elif "test" in self.data_path and "joint_motion" in self.data_path:
test_data_file = test_data_joint_motion_files[index]
filename = test_data_file
f = h5py.File(test_data_joint_motion_path + test_data_file, "r")
data_numpy = f[list(f.keys())[-1]][:]
label = int(filename[filename.index("A") + 1: filename.index("A") + 4]) - 1
f.close()
'''
if "train" in self.data_path and "joint_motion" in self.data_path:
train_data_file = train_data_joint_motion_files[index]
filename = train_data_file
data_numpy = np.load(train_data_joint_motion_path +
train_data_file)
label = int(
filename[filename.index("A") + 1:filename.index("A") + 4]) - 1
elif "test" in self.data_path and "joint_motion" in self.data_path:
test_data_file = test_data_joint_motion_files[index]
filename = test_data_file
data_numpy = np.load(test_data_joint_motion_path + test_data_file)
label = int(
filename[filename.index("A") + 1:filename.index("A") + 4]) - 1
data_numpy = data_numpy[:, :, :67, :]
if self.normalization:
data_numpy = (data_numpy - self.mean_map) / self.std_map
if self.random_shift:
data_numpy = tools.random_shift(data_numpy)
if self.random_choose:
data_numpy = tools.random_choose(data_numpy, self.window_size)
elif self.window_size > 0:
data_numpy = tools.auto_pading(data_numpy, self.window_size)
if self.random_move:
data_numpy = tools.random_move(data_numpy)
return data_numpy, label, index
def __getitem__(self, index):
data_numpy = self.data[index]
label = self.label[index]
data_numpy = np.array(data_numpy)
if self.normalization:
data_numpy = (data_numpy - self.mean_map) / self.std_map
if self.random_shift:
data_numpy = tools.random_shift(data_numpy)
if self.random_choose:
data_numpy = tools.random_choose(data_numpy, self.window_size)
elif self.window_size > 0:
data_numpy = tools.auto_pading(data_numpy, self.window_size)
if self.random_move:
data_numpy = tools.random_move(data_numpy)
return data_numpy, label, index
def __getitem__(self, index):
data_numpy = self.data[index]
label = self.label[index]
data_numpy = np.array(data_numpy)
if self.random_choose:
data_numpy = tools.random_choose(data_numpy, self.window_size)
if self.random_mirror:
if random.random() > self.random_mirror_p:
assert data_numpy.shape[2] == 27
data_numpy = data_numpy[:,:,flip_index,:]
if self.is_vector:
data_numpy[0,:,:,:] = - data_numpy[0,:,:,:]
else:
data_numpy[0,:,:,:] = 512 - data_numpy[0,:,:,:]
if self.normalization:
# data_numpy = (data_numpy - self.mean_map) / self.std_map
assert data_numpy.shape[0] == 3
if self.is_vector:
data_numpy[0,:,0,:] = data_numpy[0,:,0,:] - data_numpy[0,:,0,0].mean(axis=0)
data_numpy[1,:,0,:] = data_numpy[1,:,0,:] - data_numpy[1,:,0,0].mean(axis=0)
else:
data_numpy[0,:,:,:] = data_numpy[0,:,:,:] - data_numpy[0,:,0,0].mean(axis=0)
data_numpy[1,:,:,:] = data_numpy[1,:,:,:] - data_numpy[1,:,0,0].mean(axis=0)
if self.random_shift:
if self.is_vector:
data_numpy[0,:,0,:] += random.random() * 20 - 10.0
data_numpy[1,:,0,:] += random.random() * 20 - 10.0
else:
data_numpy[0,:,:,:] += random.random() * 20 - 10.0
data_numpy[1,:,:,:] += random.random() * 20 - 10.0
# if self.random_shift:
# data_numpy = tools.random_shift(data_numpy)
# elif self.window_size > 0:
# data_numpy = tools.auto_pading(data_numpy, self.window_size)
if self.random_move:
data_numpy = tools.random_move(data_numpy)
return data_numpy, label, index
def __getitem__(self, index):
data_numpy = self.data[index]
label = self.label[index]
data_numpy = np.array(data_numpy)
info_numpy = np.array(self.sample_name[index])
if self.normalization:
data_numpy = (data_numpy - self.mean_map) / self.std_map
if self.random_shift:
data_numpy = tools.random_shift(data_numpy)
if self.random_choose:
data_numpy = tools.random_choose(data_numpy, self.window_size)
elif self.window_size > 0:
data_numpy = tools.auto_pading(data_numpy, self.window_size)
if self.random_move:
data_numpy = tools.random_move(data_numpy)
data_numpy = tools.frame_delay(data_numpy, self.frame_delay)
data_numpy = tools.frame_null(data_numpy, self.frame_null)
data_numpy = tools.horizontal_flip(data_numpy, self.horizontal_flip)
return data_numpy, label, info_numpy
def __getitem__(self, index):
data_numpy = self.data[index]
label = self.label[index]
data_numpy = np.array(data_numpy)
if self.normalization:
data_numpy = (data_numpy - self.mean_map) / self.std_map
if self.random_shift:
data_numpy = tools.random_shift(data_numpy)
if self.random_choose:
data_numpy = tools.random_choose(data_numpy, self.window_size)
elif self.window_size > 0:
data_numpy = tools.auto_pading(data_numpy, self.window_size)
if self.random_move:
data_numpy = tools.random_move(data_numpy)
disps = displacementVectors(sample=data_numpy)
rel_coords = relativeCoordinates(sample=data_numpy)
data_numpy = np.concatenate([disps, rel_coords], axis=0)
return data_numpy, label, index