def load_data(self, path, load_metrics):
filename, _ = os.path.splitext(os.path.basename(path))
data = np.load(path, allow_pickle=True, encoding='latin1')['data'].item()
data_train = data['train']
data_valid = data['test']
indices_to_select = [0, 1, 4, 7, 2, 5, 8, 6, 12, 15, 17, 19, 21, 16, 18, 20]
self._data_train['2d'] = data_train["combined_2d"][:, indices_to_select, :]
self._data_train['3d'] = data_train["combined_3d_cam"][:, indices_to_select, :]*1000
self._data_valid['2d'] = data_valid["combined_2d"][:, indices_to_select, :]
self._data_valid['3d'] = data_valid["combined_3d_cam"][:, indices_to_select, :]*1000
print("Normalizing screen coordinates")
for i in tqdm(range(self._data_train['3d'].shape[0])):
self._data_train['3d'][i, :] -= self._data_train['3d'][i, 0]
if self.center_2d:
self._data_train['2d'][i, :] -= self._data_train['2d'][i, 0]
self.plot_random()
if not load_metrics:
self.mean_2d = np.mean(self._data_train['2d'], axis=0)
self.std_2d = np.std(self._data_train['2d'], axis=0)
self.mean_3d = np.mean(self._data_train['3d'], axis=0)
self.std_3d = np.std(self._data_train['3d'], axis=0)
if not os.path.exists(os.path.join("metrics/", filename + "_metrics.npz")):
np.savez_compressed(
os.path.join("metrics/", filename + "_metrics"),
mean_2d=self.mean_2d, std_2d=self.std_2d,
mean_3d=self.mean_3d, std_3d=self.std_3d)
else:
data = np.load(load_metrics)
self.mean_2d = data['mean_2d']
self.std_2d = data['std_2d']
self.mean_3d = data['mean_3d']
self.std_3d = data['std_3d']
self._data_train['3d'] = normalize_data(self._data_train['3d'], self.mean_3d, self.std_3d, skip_root=True)
self._data_train['2d'] = normalize_data(self._data_train['2d'], self.mean_2d, self.std_2d, skip_root=self.center_2d)
for i in tqdm(range(self._data_valid['3d'].shape[0])):
self._data_valid['3d'][i, :] -= self._data_valid['3d'][i, 0]
if self.center_2d:
self._data_valid['2d'][i, :] -= self._data_valid['2d'][i, 0]
self._data_valid['3d'] = normalize_data(self._data_valid['3d'], self.mean_3d, self.std_3d, skip_root=True)
self._data_valid['2d'] = normalize_data(self._data_valid['2d'], self.mean_2d, self.std_2d, skip_root=self.center_2d)
visualize_2d(self._data_train['2d'][0, :, :], 'debug3dpw')
def load_data(self, path_train, path_valid):
data_train = np.load(path_train, allow_pickle=True)
data_valid = np.load(path_valid, allow_pickle=True)
indices_to_arrange = [0, 2, 1, 3, 5, 4, 6, 8, 7, 9, 11, 10, 12, 14, 13, 15, 17, 16, 19, 18, 21, 20, 23, 22]
indices_to_select = [0, 1, 4, 7, 2, 5, 8, 6, 12, 15, 17, 19, 21, 16, 18, 20]
self._data_train['2d'] = data_train["data_2d"][:, indices_to_arrange, :][:, indices_to_select, :]
self._data_train['3d'] = data_train["data_3d"][:, indices_to_arrange, :][:, indices_to_select, :]
self._data_valid['2d'] = data_valid["data_2d"][:, indices_to_arrange, :][:, indices_to_select, :]
self._data_valid['3d'] = data_valid["data_3d"][:, indices_to_arrange, :][:, indices_to_select, :]
# visualize_2d(self._data_train['2d'][0, :, :], 'debug_surreal')
print("Normalizing screen coordinates")
for i in tqdm(range(self._data_train['3d'].shape[0])):
self._data_train['3d'][i, :] -= self._data_train['3d'][i, 0]
if self.center_2d:
self._data_train['2d'][i, :] -= self._data_train['2d'][i, 0]
self.plot_random()
self.mean_2d = np.mean(self._data_train['2d'], axis=0)
self.std_2d = np.std(self._data_train['2d'], axis=0)
self.mean_3d = np.mean(self._data_train['3d'], axis=0)
self.std_3d = np.std(self._data_train['3d'], axis=0)
self._data_train['3d'] = normalize_data(self._data_train['3d'], self.mean_3d, self.std_3d, skip_root=True)
self._data_train['2d'] = normalize_data(self._data_train['2d'], self.mean_2d, self.std_2d, skip_root=self.center_2d)
for i in tqdm(range(self._data_valid['3d'].shape[0])):
self._data_valid['3d'][i, :] -= self._data_valid['3d'][i, 0]
if self.center_2d:
self._data_valid['2d'][i, :] -= self._data_valid['2d'][i, 0]
self._data_valid['3d'] = normalize_data(self._data_valid['3d'], self.mean_3d, self.std_3d, skip_root=True)
self._data_valid['2d'] = normalize_data(self._data_valid['2d'], self.mean_2d, self.std_2d, skip_root=self.center_2d)
def load_data(self, path, load_metrics):
filename, _ = os.path.splitext(os.path.basename(path))
indices_to_select_2d = [0, 6, 7, 8, 1, 2, 3, 12, 13, 15, 25, 26, 27, 17, 18, 19]
indices_to_select_3d = [1, 2, 3, 6, 7, 8, 12, 13, 14, 15, 17, 18, 19, 25, 26, 27]
self.cameras = cameras.load_cameras(os.path.join(path, "cameras.h5"))
TRAIN_SUBJECTS = [1, 5, 6, 7, 8]
TEST_SUBJECTS = [9, 11]
actions = ["Directions","Discussion","Eating","Greeting",
"Phoning","Photo","Posing","Purchases",
"Sitting","SittingDown","Smoking","Waiting",
"WalkDog","Walking","WalkTogether"]
trainset = self.load_3d_data(path, TRAIN_SUBJECTS, actions)
testset = self.load_3d_data(path, TEST_SUBJECTS, actions)
d2d_train, d3d_train = self.project_to_cameras(trainset, augment_count=self.augment_count)
d2d_valid, d3d_valid = self.project_to_cameras(testset)
if self.center_2d:
self._data_train['2d'] = self.root_center(np.array(d2d_train))[:, indices_to_select_2d, :]
else:
self._data_train['2d'] = np.array(d2d_train)[:, indices_to_select_2d, :]
self._data_train['3d'] = self.root_center(np.array(d3d_train))[:, indices_to_select_2d, :]
if self.center_2d:
self._data_valid['2d'] = self.root_center(np.array(d2d_valid))[:, indices_to_select_2d, :]
else:
self._data_valid['2d'] = np.array(d2d_valid)[:, indices_to_select_2d, :]
self._data_valid['3d'] = self.root_center(np.array(d3d_valid))[:, indices_to_select_2d, :]
self.plot_random()
if not load_metrics:
self.mean_3d = np.mean(self._data_train['3d'], axis=0)
self.std_3d = np.std(self._data_train['3d'], axis=0)
self.mean_2d = np.mean(self._data_train['2d'], axis=0)
self.std_2d = np.std(self._data_train['2d'], axis=0)
if not os.path.exists(os.path.join("metrics/", filename + "_metrics.npz")):
np.savez_compressed(
os.path.join("metrics/", filename + "_metrics"),
mean_2d=self.mean_2d, std_2d=self.std_2d,
mean_3d=self.mean_3d, std_3d=self.std_3d)
else:
data = np.load(load_metrics)
self.mean_2d = data['mean_2d']
self.std_2d = data['std_2d']
self.mean_3d = data['mean_3d']
self.std_3d = data['std_3d']
self._data_train['3d'] = normalize_data(self._data_train['3d'], self.mean_3d, self.std_3d, skip_root=True)
self._data_train['2d'] = normalize_data(self._data_train['2d'], self.mean_2d, self.std_2d, skip_root=self.center_2d)
self._data_valid['3d'] = normalize_data(self._data_valid['3d'], self.mean_3d, self.std_3d, skip_root=True)
self._data_valid['2d'] = normalize_data(self._data_valid['2d'], self.mean_2d, self.std_2d, skip_root=self.center_2d)
def load_data(self, path, load_metrics):
filename = os.path.splitext(os.path.basename(path))[0]
data = np.load(path, allow_pickle=True)['data'].item()
indices_to_select_2d = [
0, 1, 2, 3, 6, 7, 8, 12, 13, 15, 17, 18, 19, 25, 26, 27
]
indices_to_select_3d = [
1, 2, 3, 6, 7, 8, 12, 13, 14, 15, 17, 18, 19, 25, 26, 27
]
actions = [
"Directions", "Discussion", "Eating", "Greeting", "Phoning",
"Photo", "Posing", "Purchases", "Sitting", "SittingDown",
"Smoking", "Waiting", "WalkDog", "Walking", "WalkTogether"
]
d2d = []
d3d = []
for action in actions:
d2d.append(data[action]['2d'])
d3d.append(data[action]['3d_cam'])
d2d = np.vstack(d2d)
d3d = np.vstack(d3d)
if self.center_2d:
self._data_valid['2d'] = self.root_center(
d2d)[:, indices_to_select_2d, :]
else:
self._data_valid['2d'] = d2d[:, indices_to_select_2d, :]
self._data_valid['3d'] = self.root_center(d3d)[:,
indices_to_select_2d, :]
if not load_metrics:
self.mean_3d = np.mean(self._data_valid['3d'], axis=0)
self.std_3d = np.std(self._data_valid['3d'], axis=0)
self.mean_2d = np.mean(self._data_valid['2d'], axis=0)
self.std_2d = np.std(self._data_valid['2d'], axis=0)
if not os.path.exists(
os.path.join("metrics/", filename + "_metrics.npz")):
np.savez_compressed(os.path.join("metrics/",
filename + "_metrics"),
mean_2d=self.mean_2d,
std_2d=self.std_2d,
mean_3d=self.mean_3d,
std_3d=self.std_3d)
else:
data = np.load(load_metrics)
self.mean_2d = data['mean_2d']
self.std_2d = data['std_2d']
self.mean_3d = data['mean_3d']
self.std_3d = data['std_3d']
self._data_valid['3d'] = normalize_data(self._data_valid['3d'],
self.mean_3d,
self.std_3d,
skip_root=True)
self._data_valid['2d'] = normalize_data(self._data_valid['2d'],
self.mean_2d,
self.std_2d,
skip_root=self.center_2d)
def load_data(self, path, load_metrics):
filename, _ = os.path.splitext(os.path.basename(path))
data = np.load(path, allow_pickle=True,
encoding='latin1')['data'].item()
data_train = data['train']
data_valid = data['test']
indices_to_select = [
0, 24, 25, 26, 29, 30, 31, 2, 5, 6, 7, 17, 18, 19, 9, 10, 11
]
indices_to_sort = [
0, 4, 5, 6, 1, 2, 3, 7, 8, 10, 14, 15, 16, 11, 12, 13
]
self._data_train['2d'] = data_train[
"2d_projected"][:, indices_to_select, :][:, indices_to_sort, :]
self._data_train['3d'] = data_train[
"3d"][:, indices_to_select, :][:, indices_to_sort, :] * 1000
# print(self._data_train['3d'][0, :, :])
self._data_valid['2d'] = data_valid[
"2d_projected"][:, indices_to_select, :][:, indices_to_sort, :]
self._data_valid['3d'] = data_valid[
"3d"][:, indices_to_select, :][:, indices_to_sort, :] * 1000
for i in range(self._data_train['3d'].shape[0]):
self._data_train['3d'][i, :] -= self._data_train['3d'][i, 0]
if self.center_2d:
self._data_train['2d'][i, :] -= self._data_train['2d'][i, 0]
for i in range(self._data_valid['3d'].shape[0]):
self._data_valid['3d'][i, :] -= self._data_valid['3d'][i, 0]
if self.center_2d:
self._data_valid['2d'][i, :] -= self._data_valid['2d'][i, 0]
self.plot_random()
if not load_metrics:
self.mean_2d = np.mean(self._data_train['2d'], axis=0)
self.std_2d = np.std(self._data_train['2d'], axis=0)
self.mean_3d = np.mean(self._data_train['3d'], axis=0)
self.std_3d = np.std(self._data_train['3d'], axis=0)
if not os.path.exists(
os.path.join("metrics/", filename + "_metrics.npz")):
np.savez_compressed(os.path.join("metrics/",
filename + "_metrics"),
mean_2d=self.mean_2d,
std_2d=self.std_2d,
mean_3d=self.mean_3d,
std_3d=self.std_3d)
else:
data = np.load(load_metrics)
self.mean_2d = data['mean_2d']
self.std_2d = data['std_2d']
self.mean_3d = data['mean_3d']
self.std_3d = data['std_3d']
self._data_train['3d'] = normalize_data(self._data_train['3d'],
self.mean_3d,
self.std_3d,
skip_root=True)
self._data_train['2d'] = normalize_data(self._data_train['2d'],
self.mean_2d,
self.std_2d,
skip_root=self.center_2d)
self._data_valid['3d'] = normalize_data(self._data_valid['3d'],
self.mean_3d,
self.std_3d,
skip_root=True)
self._data_valid['2d'] = normalize_data(self._data_valid['2d'],
self.mean_2d,
self.std_2d,
skip_root=self.center_2d)
visualize_2d(self._data_train['2d'][0, :, :], 'debug3dpw')