-
Notifications
You must be signed in to change notification settings - Fork 13
/
Copy pathtrain.py
158 lines (138 loc) · 6.68 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
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
#!/usr/bin/python3
# -*- coding: utf-8 -*-
# @Author : BADBADBADBOY
# @File : train.py
# @Time : 2024/8/20 21:22
import os
import time
import torch
from dataload.dataload import DataLoad
import random
import argparse
import numpy as np
import torch.backends.cudnn as cudnn
import torch.distributed as dist
from model.loss import CtdetLoss
import torch.optim as optim
from utils.utils import AverageMeter
from config import *
from optimizer import *
def init_seeds(seed=0, cuda_deterministic=True):
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
# Speed-reproducibility tradeoff https://pytorch.org/docs/stable/notes/randomness.html
if cuda_deterministic: # slower, more reproducible
cudnn.deterministic = True
cudnn.benchmark = False
else: # faster, less reproducible
cudnn.deterministic = False
cudnn.benchmark = True
def trainProgram(opt):
local_rank = opt.local_rank
init_seeds(local_rank)
dist.init_process_group(backend='nccl')
torch.cuda.set_device(local_rank)
assert BASE_MODEL in MODEL_SUPPORT,"not support this model !!!"
if BASE_MODEL == "resnet":
from model.models_resnet import CardDetectionCorrectionModel
model = CardDetectionCorrectionModel(num_layers = NUM_LAYER, need_ftype=NEED_FTYPE)
elif BASE_MODEL == "lcnet":
from model.models_lcnet import CardDetectionCorrectionModel
model = CardDetectionCorrectionModel(ratio = MODEL_RATIO, need_ftype=NEED_FTYPE)
elif BASE_MODEL == "replcnet":
from model.models_replcnet import CardDetectionCorrectionModel
model = CardDetectionCorrectionModel(ratio=MODEL_RATIO, need_ftype=NEED_FTYPE)
if OPTIM_METHOD == "Adam":
optimizer = AdamDecay(OPTIM_CONFIG,model.parameters())
else:
optimizer = SGDDecay(OPTIM_CONFIG, model.parameters())
criterion = CtdetLoss(HM_WEIGHT,WH_WEIGHT,OFFSET_WEIGHT,ANGLE_WEIGHT)
if LOAD_PRE_MODEL:
if LOAD_TYPE == "resnet18":
pre_dict = torch.load(PRE_MODEL_PATH)
new_pre_dict = {}
for key in pre_dict.keys():
if "layer4" in key:
new_pre_dict[key] = model.state_dict()[key]
else:
new_pre_dict[key] = pre_dict[key]
model.load_state_dict(new_pre_dict,strict=False)
elif LOAD_TYPE == "modelscope" and BASE_MODEL=="resnet":
model.load_state_dict(torch.load(PRE_MODEL_PATH)['state_dict'])
else:
model.load_state_dict(torch.load(PRE_MODEL_PATH))
train_dataset = DataLoad(img_dir,gt_dir,MAX_OBJS, INPUT_H, INPUT_W)
train_sampler = torch.utils.data.distributed.DistributedSampler(train_dataset)
train_loader = torch.utils.data.DataLoader(
train_dataset,
batch_size=BATCH_SIZE,
sampler=train_sampler,
num_workers=NUM_WORKER,
pin_memory=True,
drop_last=True
)
model = model.cuda(local_rank)
model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[local_rank], find_unused_parameters=True)
criterion = criterion.cuda(local_rank)
if dist.get_rank() == 0:
print(model)
if not os.path.exists(SAVE_PATH):
os.mkdir(SAVE_PATH)
fid = open(os.path.join(SAVE_PATH, 'loss_{}.txt'.format(BASE_MODEL)),'w+', encoding='utf-8')
total_iter = len(train_loader) * NUM_EPOCH
OPTIM_CONFIG['optimizer']['n_epoch'] = total_iter
for epoch in range(NUM_EPOCH):
train_sampler.set_epoch(epoch)
current_lr = optimizer.param_groups[0]['lr']
epoch_ave_total = AverageMeter()
epoch_ave_hm = AverageMeter()
epoch_ave_wh = AverageMeter()
epoch_ave_reg = AverageMeter()
epoch_ave_cls = AverageMeter()
data_time_record = AverageMeter()
model_time_record = AverageMeter()
data_s = time.time()
for iter_id, batch in enumerate(train_loader):
data_e = time.time()
train_iter = (len(train_loader) * epoch + iter_id)
for key in batch.keys():
batch[key] = batch[key].cuda(local_rank)
model_s = time.time()
outputs = model(batch['input'])
loss, loss_stats = criterion(outputs, batch)
optimizer.zero_grad()
loss.backward()
optimizer.step()
model_e = time.time()
model_time_record.update(model_e - model_s)
data_time_record.update(data_e - data_s)
epoch_ave_total.update(loss_stats['loss'].item())
epoch_ave_hm.update(loss_stats['hm_loss'].item())
epoch_ave_wh.update(loss_stats['wh_loss'].item())
epoch_ave_reg.update(loss_stats['off_loss'].item())
epoch_ave_cls.update(loss_stats['cls_loss'].item())
if train_iter % SHOW_STEP == 0 and dist.get_rank() == 0:
print_log = "Train iters:{}/{}\t".format(train_iter,total_iter)
for key in loss_stats.keys():
print_log += "{}:{}".format(key,"%.4f"%loss_stats[key].item())
print_log += '\t'
print_log += "data_time:{}".format("%.3f"%data_time_record.avg)
print_log += "\tmodel_time:{}".format("%.3f"%model_time_record.avg)
print_log += "\tLr:{}".format("%.8f"%current_lr)
print(print_log)
if train_iter % SAVE_ITERS == 0 and train_iter != 0 and dist.get_rank() == 0:
torch.save(model.state_dict(),"{}/model_{}.pth".format(SAVE_PATH,train_iter))
print("**************************************************************************")
print("Save Iter {}/{}\t loss:{}\thm_loss:{}\twh_loss:{}\toff_loss:{}\tcls_loss:{}".format(train_iter,total_iter,"%.4f"%epoch_ave_total.avg,"%.4f"%epoch_ave_hm.avg,"%.4f"%epoch_ave_wh.avg,"%.4f"%epoch_ave_reg.avg,"%.4f"%epoch_ave_cls.avg))
print("Save {} success!!".format("./model_save/model_{}.pth".format(train_iter)))
print("**************************************************************************")
fid.write("Save Iter {}/{}\t loss:{}\thm_loss:{}\twh_loss:{}\toff_loss:{}\tcls_loss:{}".format(train_iter,total_iter,"%.4f"%epoch_ave_total.avg,"%.4f"%epoch_ave_hm.avg,"%.4f"%epoch_ave_wh.avg,"%.4f"%epoch_ave_reg.avg,"%.4f"%epoch_ave_cls.avg)+'\n')
fid.flush()
adjust_learning_rate_poly(OPTIM_CONFIG, optimizer, train_iter)
data_s = time.time()
if __name__=="__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--local_rank', type=int, default=-1, help='DDP parameter, do not modify')
opt = parser.parse_args()
trainProgram(opt)