/
train.py
157 lines (130 loc) · 5.85 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
import os
from os.path import join
import sys
import shutil
import argparse
import utils
import torch
from torch.utils.data import Dataset, DataLoader
from torch.optim import Adam
from torchsummary import summary
import math
import pdb
import evaluation
import readers.transform as transform
import matplotlib.pyplot as plt
import numpy as np
import readers.cityscapes_reader as city_reader
def get_args():
parser = argparse.ArgumentParser()
parser.add_argument('--model', type=str)
parser.add_argument('--reader', type=str)
parser.add_argument('--reshape-size', type=int, default=1)
parser.add_argument('--crop-size', type=int, default=1)
parser.add_argument('--save-outputs', type=int, default=0)
parser.add_argument('--save-name', type=str, default='')
parser.add_argument('--data-path', type=str, default='./data/')
parser.add_argument('--batch-size', type=int, default=1)
parser.add_argument('--num-workers', type=int, default=1)
parser.add_argument('--num-epochs', type=int, default=1)
parser.add_argument('--pretrained', type=int, default=1)
parser.add_argument('--odin', type=int, default=0)
return parser.parse_args()
def prepare_for_saving():
global save_dir
save_dir = join('./results', args.save_name)
if os.path.exists(save_dir):
shutil.rmtree(save_dir)
split_classes = ['params']
for class_name in split_classes:
os.makedirs(join(save_dir, class_name), exist_ok=True)
log_file = open(join(save_dir, 'log.txt'), 'w')
sys.stdout = utils.Logger(sys.stdout, log_file)
def evaluate_segmentation(model, loader):
conf_mats = {}
conf_mats['seg'] = torch.zeros((num_classes, num_classes), dtype=torch.int64).cuda()
conf_mats['seg_w_outlier'] = torch.zeros((num_classes+1, num_classes+1), dtype=torch.int64).cuda()
log_interval = max(len(loader) // 5, 1)
for step, batch in enumerate(loader):
try:
evaluation.segment_image(model, batch, args, conf_mats, ood_id, num_classes)
except Exception as e:
print('failed on image: {}'.format(batch['name'][0]))
print('error: {}'.format(e))
print(traceback.format_exc())
if step % log_interval == 0:
print('step {} / {}'.format(step, len(loader)))
print('\nSegmentation:')
conf_mats['seg'] = conf_mats['seg'].cpu().numpy()
evaluation.compute_errors(conf_mats['seg'], 'Validation', class_info, nc=num_classes, verbose=True)
print('\nSegmentation with confidence:')
conf_mats['seg_w_outlier'] = conf_mats['seg_w_outlier'].cpu().numpy()
evaluation.compute_errors(conf_mats['seg_w_outlier'], 'Validation', class_info, nc=num_classes)
seed = 314159
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
np.random.seed(seed)
args = get_args()
if args.save_outputs:
prepare_for_saving()
net_model = utils.import_module('net_model', args.model)
reader = utils.import_module('reader', args.reader)
class_info = reader.DatasetReader.class_info
num_classes = reader.DatasetReader.num_classes
ignore_id = reader.DatasetReader.ignore_id
ood_id = reader.DatasetReader.ood_id
model = net_model.build(pretrained=True, args=args, num_classes=num_classes, ignore_id=ignore_id)
model = model.cuda()
summary(model,(3, 512,512))
train_dataset = reader.DatasetReader(args, 'train', train=True)
val_dataset = city_reader.DatasetReader(args, 'val')
#train_eval_dataset = reader.DatasetReader(args, 'train')
train_loader = DataLoader(train_dataset, batch_size=args.batch_size, shuffle=True, num_workers=args.num_workers, pin_memory=True, drop_last=True)
val_loader = DataLoader(val_dataset, batch_size=1, shuffle=False, num_workers=1, pin_memory=True)
#train_eval_loader = DataLoader(train_eval_dataset, batch_size=1, shuffle=False, num_workers=1, pin_memory=True)
fine_tune = []
fine_tune.extend(model.feature_extractor.backbone.features.parameters())
print(len(fine_tune))
random_init = []
random_init.extend(model.feature_extractor.upsample_layers.parameters())
random_init.extend(model.feature_extractor.spp.parameters())
random_init.extend(model.logits.parameters())
#random_init.extend(model.aux_logits.parameters())
optimizer = Adam([{'params':fine_tune, 'lr_factor':4},
{'params':random_init, 'lr_factor':1}],
lr=4e-4, eps=1e-5, weight_decay=1e-4, amsgrad=True)
optimizer.zero_grad()
for epoch in range(0, args.num_epochs):
lr = 1e-10+(4e-4-1e-10)*(1 + math.cos(epoch/args.num_epochs * math.pi)) / 2
print('epoch: {}, lr: {}'.format(epoch, lr))
for param_group in optimizer.param_groups:
param_group['lr'] = lr / param_group['lr_factor']
for param_group in optimizer.param_groups:
lr = param_group['lr']
print('LR = ', lr)
model = model.train()
log_interval = max(len(train_loader) // 10, 1)
for step, batch in enumerate(train_loader):
#for img_ind in range(batch['image'].shape[0]):
# plt.subplot(1,3,1)
# plt.imshow(transform.denormalize(batch['image'][img_ind], (batch['mean'][0][0], batch['mean'][1][0], batch['mean'][2][0]), (batch['std'][0][0], batch['std'][1][0], batch['std'][2][0])))
# plt.subplot(1,3,2)
# plt.imshow(batch['labels'][img_ind,:,:].numpy())
# plt.subplot(1,3,3)
# plt.imshow(batch['labels_ood'][img_ind,:,:].numpy())
# plt.show()
#pdb.set_trace()
optimizer.zero_grad()
loss = model.get_loss(batch)
loss.backward()
optimizer.step()
if step % log_interval == 0:
print('step {} / {}, loss: {}'.format(step, len(train_loader), loss.item()))
torch.save(model.state_dict(), join(
save_dir, 'params', 'epoch_{}.pt'.format(epoch)))
with torch.no_grad():
model = model.eval()
evaluate_segmentation(model, val_loader)
#with torch.no_grad():
# model = model.eval()
# evaluate_segmentation(model, train_eval_loader)