-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
133 lines (114 loc) · 5.07 KB
/
train.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
# -*- coding: utf-8 -*-
# @Time : 2021/3/30
# @Author : Xiaoyu Liu
# @Email : liuxyu@mail.ustc.edu.cn
# @Software: PyCharm
import torch
import argparse
import numpy as np
from solver import Solver
from model.unet2d_residual import ResidualUNet2D
from model.Unet_EGNN import unet_egnn
from utils.utils import log_args
from data.dataset import CVPPP
from torch.utils.data import DataLoader
from utils.logger import Log
import os
parser = argparse.ArgumentParser()
parser.add_argument('--local_rank', type=int, default=0,help='node rank for distributed training')
parser.add_argument("-b", "--batch_size",type=int,default=1)
parser.add_argument("-g", "--gpu_nums",type=int,default=1)
parser.add_argument("-e", "--epochs",type=int,default=1000)
parser.add_argument("-r", "--lr",type=float,default=1e-3)
parser.add_argument("-p", "--lr_patience",type=int,default=30)
parser.add_argument("-n", "--network",type=str,default="unet_egnn(3,[16,32,64,128,256],3,args)")
parser.add_argument("-t", "--loss_type",type=str,default="BCE_loss")
parser.add_argument("-d", "--data_dir",type=str,default="/braindat/lab/liuxy/superpixel/A1")
parser.add_argument("-l", "--logs_dir",type=str,default="./log")
parser.add_argument("-c", "--ckps_dir",type=str,default="./ckp")
# parser.add_argument("-s", "--resample",type=tuple,default=(1, 0.25, 0.25),help="resample rate:(z,h,w)")
parser.add_argument("-w", "--weight_rate",type=list,default=[10,1])
parser.add_argument("-x", "--resume",type=bool,default=False)
parser.add_argument("-y", "--resume_path",type=str,default="./ckp/checkpoint-epoch530.pth")
# parser.add_argument("-z", "--tolerate_shape",type=tuple,default=(192, 384, 384))
#spixel
parser.add_argument('--train_img_height', '-t_imgH', default = 448, type=int, help='img height')
parser.add_argument('--train_img_width', '-t_imgW', default = 448, type=int, help='img width')
parser.add_argument('--input_img_height', '-v_imgH', default = 448, type=int, help='img height')
parser.add_argument('--input_img_width', '-v_imgW', default = 448, type=int, help='img width')
#embedding
parser.add_argument("-a", "--alpha",type=int,default=1)
parser.add_argument("-be", "--beta",type=int,default=1)
parser.add_argument("-ga", "--gama",type=int,default=0.001)
#EGNN
#loss rate
parser.add_argument("-ls", "--loss_spixel",type=int,default=5) # for affinity
parser.add_argument("-le", "--loss_embedding",type=int,default=1)
parser.add_argument("-lb", "--loss_binary",type=int,default=100)
parser.add_argument("-lg", "--loss_gnn",type=int,default=10)
args = parser.parse_args()
SEED = 123
torch.manual_seed(SEED)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = True
np.random.seed(SEED)
log = Log()
if __name__ == '__main__':
#DDP
torch.cuda.set_device(args.local_rank)
torch.distributed.init_process_group(
'nccl',
init_method='env://'
)
device = torch.device(f'cuda:{args.local_rank}')
gpus = args.gpu_nums
model = eval(args.network)
#load pretrained model
model_path = os.path.join(r'./checkpoint-epoch195.pth')
net_dict = model.state_dict()
pretrain = torch.load(model_path)
pretrain_dict = {'unet.'+k: v for k, v in pretrain['state_dict'].items() if 'unet.'+k in net_dict.keys()}
net_dict.update(pretrain_dict)
model.load_state_dict(net_dict)
for k, v in model.named_parameters():
# print(k)
if 'unet.inconv' in k or 'unet.down' in k or '_spix' in k:
v.requires_grad = False
print('following parameters are training:')
for name, param in model.named_parameters():
if param.requires_grad:
print(name)
criterion = args.loss_type
metric = "dc_score"
batch_size = args.batch_size
epochs = args.epochs
lr = args.lr
trainset = CVPPP(dir=args.data_dir,mode="train",size=args.train_img_height)
train_sampler = torch.utils.data.distributed.DistributedSampler(trainset)
train_loader = DataLoader(trainset, batch_size=batch_size, shuffle=False, sampler=train_sampler)
valset = CVPPP(dir=args.data_dir,mode="validation",size=args.input_img_height)
val_sampler = torch.utils.data.distributed.DistributedSampler(valset)
val_loader = DataLoader(valset,batch_size=batch_size,shuffle=False,sampler=val_sampler)
logs_dir = args.logs_dir
patience = args.lr_patience
checkpoint_dir = args.ckps_dir
# scale = args.resample
weight = args.weight_rate
resume = args.resume
resume_path = args.resume_path
# tolerate_shape = args.tolerate_shape
#embedding
alpha = args.alpha
beta = args.beta
gama = args.gama
#spixel
le = args.loss_embedding
ls = args.loss_spixel
lb = args.loss_binary
log_args(args, log)
solver = Solver(gpus=gpus,model=model,criterion=criterion,metric=metric,batch_size=batch_size,
epochs=epochs,lr=lr,trainset=trainset,valset=valset,train_loader=train_loader,
val_loader=val_loader,logs_dir=logs_dir,patience=patience,
checkpoint_dir=checkpoint_dir,weight=weight, resume=resume,resume_path=resume_path,
log=log,args = args)
solver.train()