-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_vipnet_finetune.py
176 lines (150 loc) · 7.58 KB
/
train_vipnet_finetune.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
from __future__ import print_function
import sys
import argparse
import random
import torch.backends.cudnn as cudnn
import torch.optim as optim
# import time,datetime
from tensorboardX import SummaryWriter
from dataset import *
from model import *
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
parser = argparse.ArgumentParser()
parser.add_argument('--dataRoot', type=str, default='/data/dudong/ShapeNetCore.v1', help='data root path')
parser.add_argument('--batchSize', type=int, default=16, help='input batch size')
parser.add_argument('--workers', type=int, help='number of data loading workers', default=8)
parser.add_argument('--model', type=str, default='./checkpoint/', help='model path')
parser.add_argument('--log', type=str, default='./log_vipnet_finetune/', help='log path')
parser.add_argument('--nepoch', type=int, default=20, help='number of epochs to train for')
parser.add_argument('--lr', type=float, default=2e-5, help='Initial learning rate.')
parser.add_argument('--weight_decay', type=float, default=1e-6, help='Weight decay (L2 loss on parameters).')
parser.add_argument('--cat', type=str, default='03001627')
parser.add_argument('--cuda', type=str, default='0')
opt = parser.parse_args()
# cat_set: '03001627', '02691156', '02958343', '04090263', '04256520', '04379243'
# '04090263'(rifle) need less training epochs(10)
os.environ["CUDA_VISIBLE_DEVICES"] = opt.cuda
opt.manualSeed = random.randint(1, 10000) # fix seed
random.seed(opt.manualSeed)
torch.manual_seed(opt.manualSeed)
# Creat train/val dataloader
dataset = ShapeNet(img_root=os.path.join(opt.dataRoot, 'renderingimg'),
vox_root=os.path.join(opt.dataRoot, 'vox64'),
pc_root=os.path.join(opt.dataRoot, 'pc'),
pc_occ_root=os.path.join(opt.dataRoot, 'pc_occ'),
filelist_root=os.path.join(opt.dataRoot, 'train_val_test_list'),
cat=opt.cat, mode='train')
dataloader = torch.utils.data.DataLoader(dataset, batch_size=opt.batchSize,
shuffle=True, num_workers=int(opt.workers))
dataset_val = ShapeNet(img_root=os.path.join(opt.dataRoot, 'renderingimg'),
vox_root=os.path.join(opt.dataRoot, 'vox64'),
pc_root=os.path.join(opt.dataRoot, 'pc'),
pc_occ_root=os.path.join(opt.dataRoot, 'pc_occ'),
filelist_root=os.path.join(opt.dataRoot, 'train_val_test_list'),
cat=opt.cat, mode='val')
dataloader_val = torch.utils.data.DataLoader(dataset_val, batch_size=opt.batchSize,
shuffle=False, num_workers=int(opt.workers))
len_dataset = len(dataset)
len_dataset_val = len(dataset_val)
print('training set num', len_dataset)
print('validation set num', len_dataset_val)
# create path
model_path = os.path.join(opt.model, opt.cat)
log_path = os.path.join(opt.log, opt.cat)
if not os.path.exists(model_path):
os.makedirs(model_path)
if not os.path.exists(log_path):
os.makedirs(log_path)
logger = SummaryWriter(log_path)
cudnn.benchmark = True
# Create network
network = SVRNetworkOccGuide()
network.cuda()
# load pre-trained models
network.img2vox.load_state_dict(torch.load(os.path.join(model_path, 'img2vox64.pt')))
print('loaded pre-trained img2vox model.')
network.img2pc.load_state_dict(torch.load(os.path.join(model_path, 'img2pc.pt')))
print('loaded pre-trained img2pc model.')
network.occnet.load_state_dict(torch.load(os.path.join(opt.model, 'occnet.pt')))
print('loaded pre-trained occnet model.')
network.voxrefine.load_state_dict(torch.load(os.path.join(model_path, 'vox64_refine.pt')))
print('loaded pre-trained voxrefine model.')
# Create Loss Module
sys.path.append('./utils/chamferdistance/')
import dist_chamfer as ext
dist_chamfer = ext.chamferDist()
criterion_bce_init = nn.BCEWithLogitsLoss(pos_weight=torch.FloatTensor([5.]).cuda())
criterion_bce_ref = nn.BCEWithLogitsLoss(pos_weight=torch.FloatTensor([2.]).cuda())
criterion_bce = nn.BCEWithLogitsLoss()
# Create optimizer
# optimizer = optim.Adam(network.parameters(), lr=opt.lr, weight_decay=opt.weight_decay)
optimizer = optim.Adam(network.parameters(), lr=opt.lr)
it_step = 0
for epoch in range(1, opt.nepoch+1):
# TRAIN MODE
network.train()
for i, data in enumerate(dataloader, 0):
it_step += 1
optimizer.zero_grad()
img, vox_gt, pc_gt, occ_data, name, view_id = data
img = img.cuda()
vox_gt = vox_gt.cuda()
pc_gt = pc_gt.cuda()
occ_pc = occ_data[None].cuda()
occ_gt = occ_data['occ'].cuda()
vox_pre, vox_pre_sigmoid, pc_pre, occ_pre, _, vox_refine_pre, vox_refine_pre_sigmoid = network(img, occ_pc)
# loss
vox_init_loss = criterion_bce_init(vox_pre, vox_gt)
dist1, dist2 = dist_chamfer(pc_gt, pc_pre)
cd_loss = torch.mean(dist1) + torch.mean(dist2)
occ_loss = criterion_bce(occ_pre, occ_gt)
vox_ref_loss = criterion_bce_ref(vox_refine_pre, vox_gt)
loss = vox_init_loss + 0.5 * cd_loss + occ_loss + 3 * vox_ref_loss
loss.backward()
optimizer.step()
if it_step % 10 == 0:
# logs
logger.add_scalar('train/vox_init_loss', vox_init_loss, it_step)
logger.add_scalar('train/cd_loss', cd_loss, it_step)
logger.add_scalar('train/occ_loss', occ_loss, it_step)
logger.add_scalar('train/vox_ref_loss', vox_ref_loss, it_step)
logger.add_scalar('train/total_loss', loss, it_step)
print('[%d: %d/%d] train vox_init_loss: %f, cd_loss: %f, vox_ref_loss: %f, occ_loss: %f, total_loss: %f' %
(epoch, i, len_dataset / opt.batchSize, vox_init_loss.item(), cd_loss.item(), vox_ref_loss.item(),
occ_loss.item(), loss.item()))
# #VALIDATION
# network.eval()
# with torch.no_grad():
# for i, data in enumerate(dataloader_val, 0):
# img, vox, pc, occ_data, name, view_id = data
# img = img.cuda()
# vox = vox.cuda()
# pc = pc.cuda()
# occ_p = occ_data[None].cuda()
# occ_val = occ_data['occ'].cuda()
#
# vox_pre, vox_pre_sigmoid, p_pre, _, _, occ_val_pre, _, vox_refine_pre, vox_refine_pre_sigmoid = network(img, occ_p)
# vox_init_loss = criterion_bce(vox_pre, vox)
# dist1, dist2 = dist_chamfer(pc, p_pre)
# cd_loss = (torch.mean(dist1) + torch.mean(dist2)) * 10.
# occ_loss = criterion_bce(occ_val_pre, occ_val)
# vox_ref_loss = criterion_bce(vox_refine_pre, vox)
#
# loss = vox_init_loss + cd_loss + 2 * occ_loss + 2 * vox_ref_loss
#
# logger.add_scalar('val/vox_init_loss', vox_init_loss, it_step)
# logger.add_scalar('val/vox_ref_loss', vox_ref_loss, it_step)
# logger.add_scalar('val/cd_loss', cd_loss, it_step)
# logger.add_scalar('val/occ_loss', occ_loss, it_step)
# logger.add_scalar('val/total_loss', loss, it_step)
#
# print('[%d: %d/%d] val vox_init_loss: %f, vox_ref_loss: %f, cd_loss: %f, occ_loss: %f, total_loss: %f' %
# (epoch, i, len_dataset / opt.batchSize, vox_init_loss.item(), vox_ref_loss.item(),
# cd_loss.item(), occ_loss.item(), loss.item()))
# break # just validate one batch
torch.save(network.state_dict(), os.path.join(model_path, 'vipnet_finetune.pt'))
print('save model succeeded!')
# if epoch % 5 == 0:
# torch.save(network.state_dict(), os.path.join(model_path, 'vipnet_finetune_'+str(epoch)+'.pt'))
# print('save model succeeded!')
print('Training done!')