-
Notifications
You must be signed in to change notification settings - Fork 2
/
data.py
192 lines (158 loc) · 9.15 KB
/
data.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
import numpy as np
import torch
from torch_geometric.data import InMemoryDataset, Data
import pickle
import hdf5storage
from mesh_operations import get_vert_connectivity
def get_saga_train_idxs(num_shapes):
outliers = np.asarray([6710, 6792, 6980]) - 1
singulars = np.asarray([1354, 1804, 2029, 4283, 4306, 4377, 4433, 5543, 5925, 6464,
9365, 9575, 9641, 9862, 10210, 10561, 11434, 11778, 11783, 13963,
14097, 14762, 15830, 15947, 15948, 15952, 16515, 16624, 16630, 16632,
16635, 19971, 20358])
remeshed = np.asarray([820, 1200, 7190, 11700, 12500, 14270, 15000, 16300, 19180, 20000]) - 1
# ---------- Split in train and test ----------
test_subj = [np.int(x) for x in (np.arange(18531, 20465) - 1)]
idxs_for_train_val = [np.int(x) for x in np.arange(0, num_shapes, 2) if (np.int(x) not in test_subj
and np.int(x) not in singulars
and np.int(x) not in outliers
and np.int(x) not in remeshed)]
idxs_for_test_full = [x for x in np.arange(0, num_shapes) if
x not in idxs_for_train_val and x not in outliers and x not in singulars]
idxs_for_test = [idxs_for_test_full[x] for x in np.arange(0, len(idxs_for_test_full), 8)]
idxs_for_val = [idxs_for_train_val[x] for x in np.arange(0, len(idxs_for_train_val), 10)]
idxs_for_train = [x for x in idxs_for_train_val if x not in idxs_for_val]
return idxs_for_train, idxs_for_val, idxs_for_test
class SagaDataset(InMemoryDataset):
def __init__(self, root_dir, saga_data_file, faces_file, type, transform=None, pre_transform=None):
self.root_dir = root_dir
self.transform = transform
self.pre_transform = pre_transform
self.saga_data_file = saga_data_file
self.faces_file = faces_file
super(SagaDataset, self).__init__(root_dir, transform, pre_transform)
if type == 'train':
data_path = self.processed_paths[0]
elif type == 'val':
data_path = self.processed_paths[1]
elif type == 'test':
data_path = self.processed_paths[2]
else:
raise Exception("train, val and test are supported data types")
norm_path = self.processed_paths[3]
norm_dict = torch.load(norm_path)
self.mean, self.std = norm_dict['mean'], norm_dict['std']
self.data, self.slices = torch.load(data_path)
if self.transform:
self.data = [self.transform(td) for td in self.data]
@property
def raw_file_names(self):
return self.saga_data_file
@property
def processed_file_names(self):
processed_files = ['training.pt', 'val.pt', 'test.pt', 'norm.pt']
return processed_files
def process(self):
all_train_data, all_val_data, all_test_data = [], [], []
data = hdf5storage.loadmat(self.saga_data_file) # Load dataset
mesh_vertices = data['meshes_noeye'].reshape(data['meshes_noeye'].shape[0], data['meshes_noeye'].shape[1],
3).astype('float32') # Vertices of the meshes
num_shapes_in_all_datasets = mesh_vertices.shape[0]
idxs_for_train, idxs_for_val, idxs_for_test = get_saga_train_idxs(num_shapes_in_all_datasets)
meshes_test_np = mesh_vertices[idxs_for_test, :, :]
meshes_val_np = mesh_vertices[idxs_for_val, :, :]
meshes_train_np = mesh_vertices[idxs_for_train, :, :]
mesh_faces_np = np.load(self.faces_file)
for i in range(len(meshes_train_np)):
train_verts = torch.Tensor(meshes_train_np[i])
train_adjacency = get_vert_connectivity(meshes_train_np[i], mesh_faces_np).tocoo()
train_edge_index = torch.Tensor(np.vstack((train_adjacency.row, train_adjacency.col)))
train_data = Data(x=train_verts, y=train_verts, edge_index=train_edge_index)
all_train_data.append(train_data)
for i in range(len(meshes_val_np)):
val_verts = torch.Tensor(meshes_val_np[i])
val_adjacency = get_vert_connectivity(meshes_val_np[i], mesh_faces_np).tocoo()
val_edge_index = torch.Tensor(np.vstack((val_adjacency.row, val_adjacency.col)))
val_data = Data(x=val_verts, y=val_verts, edge_index=val_edge_index)
all_val_data.append(val_data)
for i in range(len(meshes_test_np)):
test_verts = torch.Tensor(meshes_test_np[i])
test_adjacency = get_vert_connectivity(meshes_test_np[i], mesh_faces_np).tocoo()
test_edge_index = torch.Tensor(np.vstack((test_adjacency.row, test_adjacency.col)))
test_data = Data(x=test_verts, y=test_verts, edge_index=test_edge_index)
all_test_data.append(test_data)
mean_train = torch.Tensor(np.mean(meshes_train_np, axis=0))
std_train = torch.Tensor(np.std(meshes_train_np, axis=0))
norm_dict = {'mean': mean_train, 'std': std_train}
if self.pre_transform is not None:
if hasattr(self.pre_transform, 'mean') and hasattr(self.pre_transform, 'std'):
if self.pre_transform.mean is None:
self.pre_transform.mean = mean_train
if self.pre_transform.std is None:
self.pre_transform.std = std_train
all_train_data = [self.pre_transform(td) for td in all_train_data]
all_val_data = [self.pre_transform(td) for td in all_val_data]
all_test_data = [self.pre_transform(td) for td in all_test_data]
torch.save(self.collate(all_train_data), self.processed_paths[0])
torch.save(self.collate(all_val_data), self.processed_paths[1])
torch.save(self.collate(all_test_data), self.processed_paths[2])
torch.save(norm_dict, self.processed_paths[3])
class AttackedDataset(InMemoryDataset):
def __init__(self, root_dir, results_file_in, faces_file, type, transform=None, pre_transform=None):
self.root_dir = root_dir
self.transform = transform
self.pre_transform = pre_transform
self.results_file_in = results_file_in
self.faces_file = faces_file
super(AttackedDataset, self).__init__(root_dir, transform, pre_transform)
if type == 'source':
data_path = self.processed_paths[0]
elif type == 'target':
data_path = self.processed_paths[1]
elif type == 'adversary':
data_path = self.processed_paths[2]
else:
raise Exception("source, target and adversary are supported data types")
self.data, self.slices = torch.load(data_path)
if self.transform:
self.data = [self.transform(td) for td in self.data]
@property
def raw_file_names(self):
return self.results_file_in
@property
def processed_file_names(self):
processed_files = ['source.pt', 'target.pt', 'adversary.pt']
return processed_files
def process(self):
all_source_data, all_target_data, all_adversary_data = [], [], []
with open(self.results_file_in, 'rb') as handle:
results_dict = pickle.load(handle)
faces_np = np.load(self.faces_file)
pair_numbers = [item["pair_number"] for item in results_dict]
s_labels = [item["s_label"] for item in results_dict]
t_labels = [item["t_label"] for item in results_dict]
s_meshes = [item["s_mesh"] for item in results_dict]
t_meshes = [item["t_mesh"] for item in results_dict]
adv_meshes = [item["adv_mesh"] for item in results_dict]
for i in range(len(s_meshes)):
s_verts = torch.Tensor(s_meshes[i])
s_adjacency = get_vert_connectivity(s_meshes[i], faces_np).tocoo()
s_edge_index = torch.Tensor(np.vstack((s_adjacency.row, s_adjacency.col)))
s_data = Data(x=s_verts, y=s_verts, edge_index=s_edge_index)
all_source_data.append(s_data)
t_verts = torch.Tensor(t_meshes[i])
t_adjacency = get_vert_connectivity(t_meshes[i], faces_np).tocoo()
t_edge_index = torch.Tensor(np.vstack((t_adjacency.row, t_adjacency.col)))
t_data = Data(x=t_verts, y=t_verts, edge_index=t_edge_index)
all_target_data.append(t_data)
adv_verts = torch.Tensor(adv_meshes[i])
adv_adjacency = get_vert_connectivity(adv_meshes[i], faces_np).tocoo()
adv_edge_index = torch.Tensor(np.vstack((adv_adjacency.row, adv_adjacency.col)))
adv_data = Data(x=adv_verts, y=adv_verts, edge_index=adv_edge_index)
all_adversary_data.append(adv_data)
all_source_data = [self.pre_transform(td) for td in all_source_data]
all_target_data = [self.pre_transform(td) for td in all_target_data]
all_adversary_data = [self.pre_transform(td) for td in all_adversary_data]
torch.save(self.collate(all_source_data), self.processed_paths[0])
torch.save(self.collate(all_target_data), self.processed_paths[1])
torch.save(self.collate(all_adversary_data), self.processed_paths[2])