-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_classify.py
188 lines (154 loc) · 6.64 KB
/
train_classify.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
import torch
import torch.nn as nn
import torch.optim as optim
from tensorboardX import SummaryWriter
import numpy as np
import os
import argparse
from im2mesh import config
#from classify.img_resnet import ImgClassify_ResNet18
#from classify.depth_resnet import DepthClassify_Resnet18
#from classify.pointnet import PointcloudClassify_Pointnet
from classify.dataset import get_dataset
from classify.model import get_model
from im2mesh import data
from tqdm import tqdm
import shutil
parser = argparse.ArgumentParser(
description='Train a 3D reconstruction model.'
)
parser.add_argument('--config', type=str, default='default', help='Path to config file.')
parser.add_argument('--batch_size', type=int, default=128, help='batch_size')
parser.add_argument('--val_batch_size', type=int, default=64, help='val_batch_size')
parser.add_argument('--out_dir', type=str, default='out/classify/depthpc_world_512_origin_subdivision')
parser.add_argument('--save_every', type=int, default=1000)
parser.add_argument('--backup_every', type=int, default=4000)
parser.add_argument('--quit_after', type=int, default=20000)
parser.add_argument('--lr_drop', type=int, default=15000)
parser.add_argument('--data_parallel', type=str, default='None')
args = parser.parse_args()
if args.config != 'default':
cfg = config.load_config(args.config, 'configs/default.yaml')
cfg_path = args.config
else:
cfg = config.load_config('configs/classify/partial_pointcloud_pointnet.yaml', 'configs/default.yaml')
cfg_path = 'configs/classify/partial_pointcloud_pointnet.yaml'
out_dir = args.out_dir
batch_size = args.batch_size
save_every = args.save_every
backup_every = args.backup_every
lr_drop = args.lr_drop
quit_after = args.quit_after
input_type = cfg['data']['input_type']
dataset_root = cfg['data']['path']
train_dataset = get_dataset(dataset_root, 'train', cfg, input_type)
val_dataset = get_dataset(dataset_root, 'val', cfg, input_type)
train_loader = torch.utils.data.DataLoader(
train_dataset, batch_size=batch_size, num_workers=4, shuffle=True,
collate_fn=data.collate_remove_none,
worker_init_fn=data.worker_init_fn)
val_loader = torch.utils.data.DataLoader(
val_dataset, batch_size=args.val_batch_size, num_workers=4, shuffle=False,
collate_fn=data.collate_remove_none,
worker_init_fn=data.worker_init_fn)
is_cuda = torch.cuda.is_available()
device = torch.device("cuda" if is_cuda else "cpu")
model = get_model(input_type, cfg)
'''
if input_type == 'img':
model = ImgClassify_ResNet18(13, c_dim=c_dim, pretrained=pretrained)
elif input_type == 'img_with_depth':
model = DepthClassify_Resnet18(13, c_dim=c_dim, pretrained=pretrained, with_img=pred_with_img)
elif input_type == 'depth_pointcloud':
model = PointcloudClassify_Pointnet(13, c_dim=c_dim, depth_pointcloud_transfer=depth_pointcloud_transfer)
else:
raise NotImplementedError
'''
if args.data_parallel == 'DP':
model = torch.nn.DataParallel(model)
model = model.to(device)
if not os.path.exists(out_dir):
os.makedirs(out_dir)
shutil.copy(cfg_path, os.path.join(out_dir, 'config.yaml'))
try:
pretrained_dict = torch.load(os.path.join(out_dir,'model.pt')).state_dict()
except :
pretrained_dict = dict()
model_dict = model.state_dict()
pretrained_dict = {k: v for k, v in pretrained_dict.items() if k in model_dict}
model_dict.update(pretrained_dict)
model.load_state_dict(model_dict)
epoch = 0
it = 0
optimizer = optim.Adam(model.parameters(), lr=1e-4)
logger = SummaryWriter(os.path.join(out_dir, 'logs'))
nparameters = sum(p.numel() for p in model.parameters())
print(model)
print('Total number of parameters: %d' % nparameters)
def get_correctness():
model.eval()
correct_count = 0
total_count = 0
print('Testing:')
for val_batch in tqdm(val_loader):
data = val_batch
idxs = data['idx']
current_batch_size = idxs.shape[0]
out = model(data, device, get_count=True)
#if isinstance(out, tuple):
# out = out[0]
#class_gt = data.get('category').to(device)
#class_predict = out.max(dim=1)[1]
#print('gt:',class_gt)
#print('pred:',class_predict)
#correct_count += (class_predict == class_gt).sum().item()
correct_count += out.sum().item()
total_count += current_batch_size
print('correct_count:',correct_count)
print('total_count:',total_count)
return correct_count / total_count
max_correctness = 0.
while True:
epoch += 1
for batch in train_loader:
model.train()
optimizer.zero_grad()
loss = model(batch, device, get_loss=True)
loss = loss.mean()
loss.backward()
optimizer.step()
logger.add_scalar('train/loss', loss, it)
print('[%d] loss: %f' %( it , loss.cpu() ) )
if (it % save_every) == 0:
current_correctness = get_correctness()
print('[%d val] correctness: %f' % (it, current_correctness))
if current_correctness > max_correctness:
max_correctness = current_correctness
output_best_path = os.path.join(out_dir,'model_best.pt')
if args.data_parallel == 'DP':
torch.save(model.module.state_dict(), output_best_path)
else:
torch.save(model.state_dict(), output_best_path)
output_best_path = os.path.join(out_dir,'encoder_best.pt')
if args.data_parallel == 'DP':
torch.save(model.module.features.state_dict(), output_best_path)
else:
torch.save(model.features.state_dict(), output_best_path)
if input_type.startswith('img'):
output_best_path = os.path.join(out_dir,'resnet18_best.pt')
if args.data_parallel == 'DP':
torch.save(model.module.features.features.state_dict(), output_best_path)
else:
torch.save(model.features.features.state_dict(), output_best_path)
output_path = os.path.join(out_dir,'model.pt')
torch.save(model.state_dict(), output_path)
if (it % backup_every) == 0:
output_path = os.path.join(out_dir,'model_%d.pt' % it)
torch.save(model.state_dict(), output_path)
if it == lr_drop:
for param_group in optimizer.param_groups:
param_group['lr'] = 1e-5
if it == quit_after:
print('exiting!')
exit(3)
it += 1