-
Notifications
You must be signed in to change notification settings - Fork 3
/
feature_extractor.py
207 lines (157 loc) · 8.01 KB
/
feature_extractor.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
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
# from tools.dual_dataloader import SingleViewDataloader, MultiViewDataloader
from tools.test_dataloader import TestDataloader
import numpy as np
import os
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
import torchvision.models as models
import argparse
import torch.optim as optim
import time
import torchvision.models as models
# from corrnet import SingleViewNet
from models.dgcnn import get_graph_feature
from models.dgcnn import DGCNN
from models.pointnet_part_seg import PointnetPartSeg
# from models.pointnet_part_seg import PointNet_Part
from models.meshnet import MeshNet
from models.SVCNN import Semi3D, SingleViewNet
def extract(img_net, dgcnn, mesh_net, num_views, split, exp_name):
dataset = 'ModelNet40'
train_set = TestDataloader(dataset, num_points = 1024, num_views = 2, partition= split)
print('length of the dataset: ', len(train_set))
data_loader = torch.utils.data.DataLoader(train_set, batch_size=1, shuffle=False,num_workers=1)
start_time = time.time()
img_feat = np.zeros((len(train_set), 256*4))
pt_feat = np.zeros((len(train_set), 256*4))
mesh_feat = np.zeros((len(train_set), 256*4))
label = np.zeros((len(train_set)))
img_pairs = 18
for idx in range(img_pairs):
print('index of image pair: ', idx)
iteration = 0
for data in data_loader:
# pt, img, _, target = data
pt, img, img_v, centers, corners, normals, neighbor_index, target = data
img = Variable(img).to('cuda')
img_v = Variable(img_v).to('cuda')
#universal features
img_rfeat, img_zfeat, img_spc = img_net(img)
imgv_rfeat, imgv_zfeat, imgv_spc = img_net(img_v)
imgF = torch.cat((img_rfeat, img_zfeat, img_spc), dim=1)
imgvF = torch.cat((imgv_rfeat, imgv_zfeat, imgv_spc), dim=1)
img_feat[iteration,:] = img_feat[iteration,:] + imgF.data.cpu().numpy() + imgvF.data.cpu().numpy()
# print(iteration)
iteration = iteration + 1
if iteration % 400 == 0:
print('iteration: ', iteration)
iteration = 0
for data in data_loader:
# pt, img, _, target = data
pt, img, img_v, centers, corners, normals, neighbor_index, target = data
target = target[:,0]
# img = Variable(img).to('cuda')
pt = Variable(pt).to('cuda')
centers = Variable(torch.cuda.FloatTensor(centers.cuda()))
corners = Variable(torch.cuda.FloatTensor(corners.cuda()))
normals = Variable(torch.cuda.FloatTensor(normals.cuda()))
neighbor_index = Variable(torch.cuda.LongTensor(neighbor_index.cuda()))
target = Variable(target).to('cuda')
pt = pt.permute(0,2,1)
#This part is for domian spcific features
# cloud_feat, _ = dgcnn(pt)
# feat, _ = img_net(img)
# # feat = img_net(img)
# # feat = feat[:,:,0,0]
# # # # print(feat.size())
# m_feat, _ = mesh_net(centers, corners, normals, neighbor_index)
#universal features
cloud_rfeat, cloud_zfeat, cloud_spc = dgcnn(pt)
# img_rfeat, img_zfeat, img_spc = img_net(img)
mesh_rfeat, mesh_zfeat, mesh_specific = mesh_net(centers, corners, normals, neighbor_index)
# print(cloud_rfeat.size(), cloud_zfeat.size(), cloud_spc.size())
cloudF = torch.cat((cloud_rfeat, cloud_zfeat, cloud_spc), dim=1)
# imgF = torch.cat((img_rfeat, img_zfeat, img_spc), dim=1)
meshF = torch.cat((mesh_rfeat, mesh_zfeat, mesh_specific), dim=1)
# m_feat = m_feat / torch.norm(m_feat)
# cloud_feat = cloud_feat / torch.norm(cloud_feat)
# feat = feat / torch.norm(feat)
# print(cloud_feat.size(), img_feat.size())
# img_feat[iteration,:] = imgF.data.cpu().numpy()
pt_feat[iteration,:] = cloudF.data.cpu().numpy()
mesh_feat[iteration,:] = meshF.data.cpu().numpy()
label[iteration] = target.data.cpu().numpy()
# print(iteration)
iteration = iteration + 1
if iteration % 1000 == 0:
print('iteration: ', iteration)
#normalize the image features
img_feat = img_feat/(2.0*img_pairs)
feature_folder = os.path.join('./extracted_features', exp_name)
if not os.path.exists(feature_folder):
os.makedirs(feature_folder)
img_feat_name = os.path.join('./extracted_features', exp_name, '%s-%s-%s_NtXent_img_feat'%(dataset, split, img_pairs))
pt_feat_name = os.path.join('./extracted_features', exp_name, '%s-%s_NtXent_cloud1024_feat'%(dataset, split))
mesh_feat_name = os.path.join('./extracted_features', exp_name, '%s-%s_NtXent_mesh_feat'%(dataset, split))
label_name = os.path.join('./extracted_features', exp_name, '%s-%s_NtXent_label'%(dataset, split))
np.save(img_feat_name, img_feat)
np.save(pt_feat_name, pt_feat)
np.save(mesh_feat_name, mesh_feat)
np.save(label_name, label)
def extract_features(args):
iterations = 10000
num_views = 2 # 1 12 80
# weights_folder = '180view-ModelNet40-pt-mesh-view-pt1024-img112-contrast-specific'
# weights_folder = '180view-ModelNet40-pt-mesh-view-pt1024-img112-contrast-NoSpecific'
weights_folder = 'ModelNet40-pt1024-mesh-img56-Xentropy-Xcontrast-PointMultiAgreement-T095-Fused-Warmup-100percent_xcenter_p10_nt_xw2_aw0_cw9_baseshare_newcenter_l2_100'
img_net = SingleViewNet(pre_trained = None)
# img_net = torch.nn.DataParallel(img_net)
img_net_name = './checkpoints/%s/%d-img_net.pkl'%(weights_folder, iterations)
img_net.load_state_dict(torch.load(img_net_name)['state_dict'])
# img_net = models.resnet18(pretrained=True)
# img_net = list(img_net.children())[:-1]
# img_net = nn.Sequential(*img_net)
# print(img_net)
dgcnn = DGCNN(args)
dgcnn_name = './checkpoints/%s/%d-pt_net.pkl'%(weights_folder, iterations)
dgcnn.load_state_dict(torch.load(dgcnn_name)['state_dict'])
# pt_net = PointnetPartSeg()
mesh_net = MeshNet()
mesh_net_name = './checkpoints/%s/%d-mesh_net.pkl'%(weights_folder, iterations)
mesh_net.load_state_dict(torch.load(mesh_net_name)['state_dict'])
# model = torch.load('./checkpoints/180view-ModelNet40-xentropy-center-pt-img/95000-head_net.pkl')
img_net = img_net.eval()
dgcnn = dgcnn.eval()
mesh_net = mesh_net.eval()
img_net = img_net.to('cuda')
dgcnn = dgcnn.to('cuda')
mesh_net = mesh_net.to('cuda')
# print('extracing features for the training split')
extract(img_net, dgcnn, mesh_net, num_views, 'train', exp_name = weights_folder)
print('extracing features for the testing split')
extract(img_net, dgcnn, mesh_net, num_views, 'test', exp_name = weights_folder)
print('------------------ Al the Features are saved ---------------------------')
if __name__ == "__main__":
# Training settings
parser = argparse.ArgumentParser(description='RGB and Point Cloud Correspondence')
parser.add_argument('--no_cuda', type=bool, default=False,
help='enables CUDA training')
parser.add_argument('--seed', type=int, default=1, metavar='S',
help='random seed (default: 1)')
parser.add_argument('--num_points', type=int, default=1024,
help='num of points to use')
parser.add_argument('--dropout', type=float, default=0.5,
help='dropout rate')
parser.add_argument('--emb_dims', type=int, default=1024, metavar='N',
help='Dimension of embeddings')
parser.add_argument('--k', type=int, default=20, metavar='N',
help='Num of nearest neighbors to use')
parser.add_argument('--model_path', type=str, default='', metavar='N',
help='Pretrained model path')
parser.add_argument('--gpu_id', type=str, default='0',
help='GPU used to train the network')
args = parser.parse_args()
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu_id
extract_features(args)