-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathmain.py
200 lines (192 loc) · 10.9 KB
/
main.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
#!/usr/bin/python
# -*- coding: utf-8 -*-
"""
======================
@author: Jaron
@time: 2021/10/6:16:27
@email: fjjth98@163.com
@description:
======================
"""
import os
import torch
import logging
import argparse
import torch.nn as nn
from tqdm import tqdm
from torch.optim import SGD
from torch.utils.data import DataLoader
from torch.optim.lr_scheduler import MultiStepLR
from torch.utils.tensorboard import SummaryWriter
from utils import SoftCrossEntropyLoss
if __name__ == '__main__':
# parse the arguments
parser = argparse.ArgumentParser('DuMLP-Pin training and evaluation script', add_help=False)
parser.add_argument('--mode', required=True, choices=['train', 'eval'])
parser.add_argument('--task', required=True, choices=['ModelNet40', 'ShapeNetPart'])
parser.add_argument('--batch-size', required=True, type=int)
parser.add_argument('--learning-rate', type=float)
parser.add_argument('--learning-rate-milestones', type=str, help='use , to separate numbers (no space)')
parser.add_argument('--gamma', type=float, default=0.1, help='gamma rate for MultiStepLR')
parser.add_argument('--momentum', type=float, default=0.9)
parser.add_argument('--weight-decay', type=float, default=1e-4)
parser.add_argument('--soft-cross-entropy', type=float, default=0., help='0. means cross entropy')
parser.add_argument('--max-epoch', type=int, default=-1, help='-1 means no stop')
parser.add_argument('--model-cls', type=str, required=True)
parser.add_argument('--model-save-dir', type=str, help='')
parser.add_argument('--model-load-path', type=str)
parser.add_argument('--data-cls', type=str, required=True)
parser.add_argument('--data-dir', type=str)
parser.add_argument('--gpu', type=int, default=-1, help='-1 = cpu')
parser.add_argument('--random-seed', type=int, default=42)
args, unparsed = parser.parse_known_args()
logger = logging.getLogger()
logger.setLevel(logging.DEBUG)
ch = logging.StreamHandler()
ch.setLevel(logging.INFO)
formatter = logging.Formatter("%(asctime)s - %(filename)s[line:%(lineno)d] - %(levelname)s: %(message)s")
ch.setFormatter(formatter)
logger.addHandler(ch)
exec('from MyModel import {} as MyModel'.format(args.model_cls))
exec('from MyDataset import {} as MyDataset'.format(args.data_cls))
torch.manual_seed(args.random_seed)
if args.gpu == -1:
use_cuda = False
device = torch.device('cpu')
else:
os.environ['CUDA_DEVICE_ORDER'] = 'PCI_BUS_ID'
os.environ['CUDA_VISIBLE_DEVICES'] = str(args.gpu)
use_cuda = True
device = torch.device('cuda')
model = MyModel().to(device)
if args.model_load_path:
model.load_state_dict(torch.load(args.model_load_path, map_location=device)['model'])
if args.model_save_dir:
if not os.path.exists(args.model_save_dir):
os.makedirs(args.model_save_dir)
writer = SummaryWriter(log_dir=args.model_save_dir)
fh = logging.FileHandler(os.path.join(args.model_save_dir, 'output.log'), mode='w')
fh.setLevel(logging.DEBUG)
fh.setFormatter(formatter)
logger.addHandler(fh)
if args.mode == 'train':
loss_fn = nn.CrossEntropyLoss().to(device) if args.soft_cross_entropy == 0. else SoftCrossEntropyLoss().to(
device)
optimizer = SGD(model.parameters(), lr=args.learning_rate, momentum=args.momentum, weight_decay=args.weight_decay)
scheduler = MultiStepLR(optimizer, milestones=[int(a) for a in args.learning_rate_milestones.split(',')], gamma=args.gamma)
if args.task == 'ModelNet40':
from utils import get_tnf, tnf2met, write_met
eval_set = MyDataset(root=args.data_dir, train=False, augment=False, cuda=use_cuda)
eval_loader = DataLoader(eval_set, batch_size=args.batch_size, shuffle=False)
if args.model_save_dir:
writer.add_graph(model, torch.randn(1, 1024, 6, device=device))
num_cats, name_cats = len(MyDataset.classes), MyDataset.classes
if args.mode == 'train':
epoch, best_met, best_epoch, record_dict = 0, -float('inf'), 0, {}
train_set = MyDataset(root=args.data_dir, train=True, augment=True, cuda=use_cuda)
train_loader = DataLoader(train_set, batch_size=args.batch_size, shuffle=True)
while epoch != args.max_epoch:
model.train()
train_tnf = torch.zeros((num_cats, 2), dtype=torch.long, device=device)
train_loss = 0.
for inputs, labels in tqdm(train_loader):
outputs = model(inputs)
loss = loss_fn(outputs, labels)
optimizer.zero_grad()
loss.backward()
optimizer.step()
train_loss += loss.item() * labels.shape[0]
get_tnf(outputs, labels, train_tnf)
model.eval()
with torch.no_grad():
eval_tnf = torch.zeros((num_cats, 2), dtype=torch.long, device=device)
eval_loss = 0.
for inputs, labels in tqdm(eval_loader):
outputs = model(inputs)
eval_loss += loss_fn(outputs, labels).item() * labels.shape[0]
get_tnf(outputs, labels, eval_tnf)
epoch += 1
scheduler.step()
my_info = 'Epoch: {:<3d}'.format(epoch)
for mode in ['eval', 'train']:
exec('record_dict["{}_loss"] = {}_loss / len({}_set)'.format(mode, mode, mode))
exec('record_dict["{}_oa"], record_dict["{}_macc"], record_dict["{}_acc"] = tnf2met({}_tnf)'.format(mode, mode, mode, mode))
my_info += '; {} OA: {:.2%}; {} mAcc: {:.2%}'.format(mode, record_dict['{}_oa'.format(mode)], mode, record_dict['{}_macc'.format(mode)])
logger.info(my_info)
if args.model_save_dir:
write_met(writer, epoch, record_dict, MyDataset.classes)
if record_dict['eval_oa'] > best_met:
best_met, best_epoch = record_dict['eval_oa'], epoch
logger.info('New best model is achieved at epoch {} with eval OA as {:.2%}'.format(best_epoch, best_met))
if args.model_save_dir:
torch.save({'epoch': epoch, 'metric': best_met, 'model': model.state_dict()}, os.path.join(args.model_save_dir, 'best_model.pth'))
logger.info('Best model is achieved at epoch {} with eval OA as {:.2%}'.format(best_epoch, best_met))
if args.model_save_dir:
torch.save({'epoch': epoch, 'metric': record_dict['eval_oa'], 'model': model.state_dict()}, os.path.join(args.model_save_dir, 'final_model.pth'))
writer.close()
else:
model.eval()
with torch.no_grad():
eval_tnf = torch.zeros((num_cats, 2), dtype=torch.long, device=device)
for inputs, labels in tqdm(eval_loader):
get_tnf(model(inputs), labels, eval_tnf)
oa, macc, _ = tnf2met(eval_tnf)
logger.info('eval OA: {:.2%}; eval mAcc: {:.2%}'.format(oa, macc))
elif args.task == 'ShapeNetPart':
from utils import get_iou, tnf2met, write_met
eval_set = MyDataset(root=args.data_dir, train=False, augment=False, cuda=use_cuda)
eval_loader = DataLoader(eval_set, batch_size=args.batch_size, shuffle=False)
num_cats, name_cats = len(MyDataset.classes), MyDataset.classes
if args.model_save_dir:
writer.add_graph(model, [torch.randn(1, 2048, 3, device=device), torch.randint(0, num_cats, (1,), device=device)])
if args.mode == 'train':
epoch, best_met, best_epoch, record_dict = 0, -float('inf'), 0, {}
train_set = MyDataset(root=args.data_dir, train=True, augment=True, cuda=use_cuda)
train_loader = DataLoader(train_set, batch_size=args.batch_size, shuffle=True)
while epoch != args.max_epoch:
model.train()
train_iou = torch.zeros((num_cats, 2), dtype=torch.float, device=device)
train_loss = 0.
for inputs, labels, pids in tqdm(train_loader):
outputs = model(inputs, labels)
loss = loss_fn(outputs.view(-1, 50), pids.view(-1))
optimizer.zero_grad()
loss.backward()
optimizer.step()
train_loss += loss.item() * labels.shape[0]
get_iou(outputs, labels, pids, train_iou)
model.eval()
with torch.no_grad():
eval_iou = torch.zeros((num_cats, 2), dtype=torch.float, device=device)
eval_loss = 0.
for inputs, labels, pids in tqdm(eval_loader):
outputs = model(inputs, labels)
eval_loss += loss_fn(outputs.view(-1, 50), pids.view(-1)).item() * labels.shape[0]
get_iou(outputs, labels, pids, eval_iou)
epoch += 1
scheduler.step()
my_info = 'Epoch: {:<3d}'.format(epoch)
for mode in ['eval', 'train']:
exec('record_dict["{}_loss"] = {}_loss / len({}_set)'.format(mode, mode, mode))
exec('record_dict["{}_piou"], record_dict["{}_ciou"], record_dict["{}_iou"] = tnf2met({}_iou)'.format(mode, mode, mode, mode))
my_info += '; {} pIoU: {:.2%}; {} cIoU: {:.2%}'.format(mode, record_dict['{}_piou'.format(mode)], mode, record_dict['{}_ciou'.format(mode)])
logger.info(my_info)
if args.model_save_dir:
write_met(writer, epoch, record_dict, MyDataset.classes)
if record_dict['eval_piou'] > best_met:
best_met, best_epoch = record_dict['eval_piou'], epoch
logger.info('New best model is achieved at epoch {} with eval pIoU as {:.2%}'.format(best_epoch, best_met))
if args.model_save_dir:
torch.save({'epoch': epoch, 'metric': best_met, 'model': model.state_dict()}, os.path.join(args.model_save_dir, 'best_model.pth'))
logger.info('Best model is achieved at epoch {} with eval pIoU as {:.2%}'.format(best_epoch, best_met))
if args.model_save_dir:
torch.save({'epoch': epoch, 'metric': record_dict['eval_piou'], 'model': model.state_dict()}, os.path.join(args.model_save_dir, 'final_model.pth'))
writer.close()
else:
model.eval()
with torch.no_grad():
eval_iou = torch.zeros((num_cats, 2), dtype=torch.float, device=device)
for inputs, labels, pids in tqdm(eval_loader):
get_iou(model(inputs, labels), labels, pids, eval_iou)
piou, ciou, _ = tnf2met(eval_iou)
logger.info('eval pIoU: {:.2%}; eval cIoU: {:.2%}'.format(piou, ciou))