-
Notifications
You must be signed in to change notification settings - Fork 36
/
Copy pathtrain_classifier.py
160 lines (130 loc) · 6.77 KB
/
train_classifier.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 open3d
import time
import copy
import numpy as np
import math
import os
import shutil
import torch
from torch.utils.tensorboard import SummaryWriter
import matplotlib
matplotlib.use('Agg')
from models.multimodal_classifier import MMClassifer, MMClassiferCoarse
from data.oxford_pc_img_pose_loader import OxfordLoader
from oxford import options
if __name__=='__main__':
opt = options.Options()
logdir = './runs/'+str(opt.version)
if os.path.isdir(logdir):
user_answer = input("The log directory %s exists, do you want to delete it? (y or n) : " % logdir)
if user_answer == 'y':
# delete log folder
shutil.rmtree(logdir)
else:
exit()
else:
os.makedirs(logdir)
# Writer will output to ./runs/ directory by default
writer = SummaryWriter(log_dir=logdir)
trainset = OxfordLoader(opt.dataroot, 'train', opt)
dataset_size = len(trainset)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=opt.batch_size, shuffle=True,
num_workers=opt.dataloader_threads, drop_last=True, pin_memory=True)
print('#training point clouds = %d' % len(trainset))
testset = OxfordLoader(opt.dataroot, 'val', opt)
testloader = torch.utils.data.DataLoader(testset, batch_size=opt.batch_size, shuffle=False,
num_workers=opt.dataloader_threads, pin_memory=True)
print('#testing point clouds = %d' % len(testset))
# create model, optionally load pre-trained model
if opt.is_fine_resolution:
model = MMClassifer(opt, writer)
else:
model = MMClassiferCoarse(opt, writer)
# model.load_model('/home/tohar/repos/point-img-feature/oxford/workspace/640x384-noCrop/checkpoints/best.pth')
best_test_accuracy = 0
for epoch in range(101):
epoch_iter = 0
for i, data in enumerate(trainloader):
pc, intensity, sn, node_a, node_b, \
P, img, K, t_ij = data
B = pc.size()[0]
iter_start_time = time.time()
epoch_iter += B
model.global_step_inc(B)
model.set_input(pc, intensity, sn, node_a, node_b,
P, img, K)
model.optimize()
if i % int(600) == 0 and i > 0:
# print/plot errors
t = (time.time() - iter_start_time) / opt.batch_size
train_loss_dict, test_loss_dict = model.get_current_errors()
train_accuracy_dict, test_accuracy_dict = model.get_current_accuracy()
model.print_loss_dict(train_loss_dict, train_accuracy_dict, t)
model.write_loss()
model.write_accuracy()
model.write_img()
model.write_pc_label(model.train_visualization['pc'],
model.train_visualization['coarse_labels'],
'coarse_labels')
model.write_pc_label(model.train_visualization['pc'],
model.train_visualization['coarse_predictions'],
'coarse_predictions')
if opt.is_fine_resolution:
model.write_classification_visualization(model.train_visualization['KP_pc_pxpy'],
model.train_visualization['coarse_predictions'],
model.train_visualization['fine_predictions'],
model.train_visualization['coarse_labels'],
model.train_visualization['fine_labels'],
t_ij)
else:
model.write_classification_visualization(model.train_visualization['KP_pc_pxpy'],
model.train_visualization['coarse_predictions'],
model.train_visualization['coarse_labels'],
t_ij)
# epoch done
test_start_time = time.time()
test_batch_sum = 0
test_loss_sum = {'loss': 0, 'coarse': 0, 'fine': 0}
test_accuracy_sum = {'coarse_accuracy': 0, 'fine_accuracy': 0}
for i, data in enumerate(testloader):
pc, intensity, sn, node_a, node_b, \
P, img, K, t_ij = data
B = pc.size()[0]
model.set_input(pc, intensity, sn, node_a, node_b,
P, img, K)
model.test_model()
_, test_loss_dict = model.get_current_errors()
_, test_accuracy = model.get_current_accuracy()
test_batch_sum += B
test_loss_sum['loss'] += B * test_loss_dict['loss']
test_loss_sum['coarse'] += B * test_loss_dict['coarse']
test_accuracy_sum['coarse_accuracy'] += B * test_accuracy['coarse_accuracy']
if opt.is_fine_resolution:
test_loss_sum['fine'] += B * test_loss_dict['fine']
test_accuracy_sum['fine_accuracy'] += B * test_accuracy['fine_accuracy']
test_loss_sum['loss'] /= test_batch_sum
test_loss_sum['coarse'] /= test_batch_sum
test_accuracy_sum['coarse_accuracy'] /= test_batch_sum
if opt.is_fine_resolution:
test_loss_sum['fine'] /= test_batch_sum
test_accuracy_sum['fine_accuracy'] /= test_batch_sum
test_persample_time = (time.time() - test_start_time) / test_batch_sum
print('Test loss and accuracy:')
model.print_loss_dict(test_loss_sum, test_accuracy_sum, test_persample_time)
# set the mean loss/accuracy to the model, so that the tensorboard visualization is correct
model.test_loss_dict = test_loss_sum
model.test_accuracy = test_accuracy_sum
# record best test loss
if test_accuracy_sum['coarse_accuracy'] > best_test_accuracy:
best_test_accuracy = test_accuracy_sum['coarse_accuracy']
print('--- best test coarse accuracy %f' % best_test_accuracy)
print('Epoch %d done.' % epoch)
if epoch % opt.lr_decay_step == 0 and epoch > 0:
model.update_learning_rate(opt.lr_decay_scale)
# save network
if epoch >= 0:
print("Saving network...")
model.save_network(model.detector, "v%s-gpu%d-epoch%d-%f.pth" % (opt.version,
opt.gpu_ids[0],
epoch,
test_accuracy_sum['coarse_accuracy']))