-
Notifications
You must be signed in to change notification settings - Fork 5
/
train.py
147 lines (119 loc) · 5.14 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
import os, sys
import yaml
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data import DataLoader
import torch.optim as optim
import numpy as np
import cv2
from tqdm import tqdm
import argparse
from tqdm import trange
from test import val
from core.dataset import LINEMOD_SO3 as LINEMOD
from core.loss import weighted_infoNCE_loss_func
from core.utils import load_checkpoint
from core.model import RetrievalNet as Model
from core.model import Sim_predictor as Predictor
np.set_printoptions(threshold=np.inf)
torch.backends.cudnn.deterministic=True
torch.backends.cudnn.enabled=True
torch.backends.cudnn.benchmark=True
torch.autograd.set_detect_anomaly(True)
torch.manual_seed(0)
np.random.seed(0)
def train_one_epoch(epoch, train_loader, model, predictor, optimizer):
model.train()
predictor.train()
train_loss = []
train_errs = []
for i, data in enumerate(train_loader):
## load data and label
src_img, ref_img, src_R, ref_R, id = data
if torch.any(id == -1):
print("Skip incorrect data")
continue
src_img = src_img.cuda()
ref_img = torch.cat(ref_img, dim=0).cuda()
src_R, ref_R = src_R.cuda(), ref_R.cuda()
id = id.cuda()
B, _, H, W = src_img.shape
## feature extraction
src_f = model(src_img)
ref_f = model(ref_img)
## similarity estimation
ref_sim = predictor(src_f, ref_f)
## loss estimation
loss = weighted_infoNCE_loss_func(ref_sim[:, B:2*B], ref_sim[:, 2*B:], ref_R[:, 1], ref_R[:, 2], src_R, id, tau=0.1)
try:
optimizer.zero_grad()
loss.backward()
optimizer.step()
except:
print("Skip incorrect data")
continue
train_loss.append(loss.item())
if i % 20 == 0:
print("\tEpoch %3d --- Iter [%d/%d] Train --- Loss: %.4f" % (epoch, i + 1, len(train_loader), loss.item()))
train_loss = np.asarray(train_loss).mean()
return train_loss
def train(cfg, device):
print(">>>>>>>>>>>>>> CREATE DATASET")
dataset_train = LINEMOD(cfg, 'train', 0)
train_loader = DataLoader(dataset_train, batch_size=cfg["TRAIN"]["BS"], shuffle=True, \
num_workers=cfg["TRAIN"]["WORKERS"], drop_last=True)
print(">>>>>>>>>> TRAINING DATA:", len(train_loader)*cfg["TRAIN"]["BS"])
print(">>>>>>>>>>>>>> CREATE NETWORK")
model = Model(cfg).to(device)
predictor = Predictor(cfg).to(device)
print(">>>>>>>>>>>>>> CREATE OPTIMIZER")
optimizer = optim.Adam([{'params': model.parameters()}, {'params': predictor.parameters()}], lr=cfg["TRAIN"]["LR"])
lrScheduler = optim.lr_scheduler.MultiStepLR(optimizer, cfg["TRAIN"]["STEP"], gamma=cfg["TRAIN"]["GAMMA"])
if not os.path.exists(cfg["TRAIN"]["WORKING_DIR"]):
os.makedirs(cfg["TRAIN"]["WORKING_DIR"])
logname = os.path.join(cfg["TRAIN"]["WORKING_DIR"], 'training_log.txt')
with open(logname, 'a') as f:
f.write('training set: ' + str(len(dataset_train)) + '\n')
if cfg["TRAIN"]["FROM_SCRATCH"] is False:
print(">>>>>>>>>>>>>> LOAD MODEL")
model, optimizer, start_epoch, best_acc = load_checkpoint(model, optimizer, cfg["TRAIN"]["WORKING_DIR"] + "checkpoint_1.pth")
predictor, _, _, _ = load_checkpoint(predictor, optimizer, cfg["TRAIN"]["WORKING_DIR"] + "checkpoint_2.pth")
else:
print(">>>>>>>>>>>>>> TRAINING FROM SCRATCH")
best_acc = 0
start_epoch = 0
print(">>>>>>>>>>>>>> START TRAINING")
for epoch in trange(start_epoch, cfg["TRAIN"]["MAX_EPOCH"]):
loss = train_one_epoch(epoch, train_loader, model, predictor, optimizer)
# update learning rate
lrScheduler.step()
if (epoch + 1) % cfg["TRAIN"]["VAL_STEP"] == 0:
res = val(cfg, model, predictor, device)
if res[0] > best_acc:
best_acc = res[0]
state_dict = {'epoch': epoch, 'state_dict': model.state_dict(),\
'optimizer': optimizer.state_dict(),
'best_acc': res[0]}
torch.save(state_dict, os.path.join(cfg["TRAIN"]["WORKING_DIR"], 'checkpoint_1.pth'))
state_dict = {'epoch': epoch, 'state_dict': predictor.state_dict(),\
'optimizer': optimizer.state_dict(),
'best_acc': res[0]}
torch.save(state_dict, os.path.join(cfg["TRAIN"]["WORKING_DIR"], 'checkpoint_2.pth'))
with open(logname, 'a') as f:
text = str('Epoch: %03d || train_loss %.4f || test_cls_acc: %.2f || test_R_acc %.2f\n' % (epoch, loss, res[1], res[0]))
f.write(text)
else:
with open(logname, 'a') as f:
text = str('Epoch: %03d || train_loss %.4f \n' % (epoch, loss))
f.write(text)
if __name__ == '__main__':
print(">>>>>>>>>>>>> Loding configuration")
with open("./objects.yaml", 'r') as load_f:
cfg = yaml.load(load_f, Loader=yaml.FullLoader)
if torch.cuda.is_available():
device = torch.device("cuda:0")
torch.cuda.set_device(device)
else:
device = torch.device("cpu")
train(cfg, device)