forked from zhengyuf/STED-gaze
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_facenet.py
164 lines (154 loc) · 6.82 KB
/
train_facenet.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
159
160
161
162
163
from models import GazeHeadNet
import numpy as np
import os
from collections import OrderedDict
import torch
import torch.nn as nn
from torch.utils.data import DataLoader, Subset
import torch.optim as optim
from torch.optim.lr_scheduler import StepLR
import logging
from core import DefaultConfig
from dataset import HDFDataset
from utils import script_init_common
logging.basicConfig(format='%(asctime)s %(message)s', level=logging.INFO)
config = DefaultConfig()
script_init_common()
batch_size = 64
num_data_loaders = 6
lr = 0.0001
train_loss_interval = 200
valid_loss_interval = 10000
save_interval = 10000
epochs = 200
test_subsample = 1.0
# Set device
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
################################################
network = faceNet()
# Transfer on the GPU before constructing and optimizer
if torch.cuda.device_count() > 1:
logging.info('Using %d GPUs!' % torch.cuda.device_count())
network = nn.DataParallel(network)
network = network.to(device)
optimizer = optim.Adam(network.parameters(), lr=lr, betas=(0.9, 0.95))
scheduler = StepLR(optimizer, step_size=50000, gamma=0.5)
from tensorboardX import SummaryWriter
tensorboard = SummaryWriter(logdir=config.save_path)
################################################
from checkpoints_manager import CheckpointsManager
saver = CheckpointsManager(network, config.save_path)
if not os.path.exists(config.save_path):
os.makedirs(config.save_path)
step = saver.load_last_checkpoint()
print("initial step: {}".format(step))
################################################
import json
with open('./gazecapture_split.json', 'r') as f:
all_gc_prefixes = json.load(f)
################################################
import losses
all_data = OrderedDict()
train_dataset = HDFDataset(hdf_file_path=config.gazecapture_file, prefixes=all_gc_prefixes['train'], num_labeled_samples=50000)
train_dataloader = DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
drop_last=True,
num_workers=num_data_loaders,
pin_memory=True,
)
all_data['gc/train'] = {'dataset': train_dataset, 'dataloader': train_dataloader}
for tag, hdf_file, is_bgr, prefixes in [
('gc/test', config.gazecapture_file, False, all_gc_prefixes['test']),
('mpi', config.mpiigaze_file, False, None),
('columbia', config.columbia_file, True, None),
('eyediap', config.eyediap_file, True, None),
]:
# Define dataset structure based on selected prefixes
dataset = HDFDataset(hdf_file_path=hdf_file,
prefixes=prefixes)
subsample = test_subsample
if subsample < (1.0 - 1e-6): # subsample if requested
dataset = Subset(dataset, np.linspace(
start=0, stop=len(dataset),
num=int(subsample * len(dataset)),
endpoint=False,
dtype=np.uint32,
))
all_data[tag] = {
'dataset': dataset,
'dataloader': DataLoader(dataset,
batch_size=batch_size,
shuffle=False,
num_workers=num_data_loaders, # args.num_data_loaders,
pin_memory=True,
),
}
def send_data_dict_to_gpu(data):
for k, v in data.items():
if isinstance(v, torch.Tensor):
data[k] = v.detach().to(device, non_blocking=True)
return data
avg_train_loss = 0.0
avg_valid_loss = 0.0
avg_train_score_g = 0.0
avg_train_score_h = 0.0
avg_valid_score_g = 0.0
avg_valid_score_h = 0.0
num_training_batches = int(len(train_dataset)/batch_size)
for epoch in range(epochs):
param_group = optimizer.param_groups[0]
train_iter = iter(train_dataloader)
for batch_i in range(num_training_batches):
input_dict = next(train_iter)
if step % save_interval == 0 and step != 0:
saver.save_checkpoint(step)
if step % train_loss_interval == 0 and step != 0:
print("step: {}, train loss: {}, train score gaze: {}, train score head: {}, lr: {}".format(
step, avg_train_loss / train_loss_interval, avg_train_score_g / train_loss_interval,
avg_train_score_h / train_loss_interval, param_group['lr']))
tensorboard.add_scalar('train/gaze', avg_train_score_g / train_loss_interval, step)
tensorboard.add_scalar('train/head', avg_train_score_h / train_loss_interval, step)
avg_train_loss = 0.0
avg_train_score_g = 0.0
avg_train_score_h = 0.0
if step % valid_loss_interval == 0 and step != 0:
network.eval()
for param in network.parameters():
param.requires_grad = False
for tag in ['columbia', 'eyediap']:
dataloader = all_data[tag]['dataloader']
for i, input_dict_ in enumerate(dataloader):
input_dict_ = send_data_dict_to_gpu(input_dict_)
gaze_h, head_h = network(input_dict_['image_a'])
score_gaze = losses.gaze_angular_loss(input_dict_['gaze_a'], gaze_h)
score_head = losses.gaze_angular_loss(input_dict_['head_a'], head_h)
loss = score_gaze + score_head
avg_valid_loss += loss
avg_valid_score_g += score_gaze
avg_valid_score_h += score_head
print("tag:{}, avg loss: {}, avg score gaze: {}, avg score head: {}"
.format(tag, avg_valid_loss / len(dataloader),avg_valid_score_g / len(dataloader),
avg_valid_score_h / len(dataloader)))
tensorboard.add_scalar('test/'+tag+'/gaze', avg_valid_score_g / len(dataloader), step)
tensorboard.add_scalar('test/'+tag+'/head', avg_valid_score_h / len(dataloader), step)
avg_valid_loss = 0.0
avg_valid_score_g = 0.0
avg_valid_score_h = 0.0
for param in network.parameters():
param.requires_grad = True
network.train()
input_dict = send_data_dict_to_gpu(input_dict)
gaze_h, head_h = network(input_dict['image_a'])
optimizer.zero_grad()
score_gaze = losses.gaze_angular_loss(input_dict['gaze_a'], gaze_h)
score_head = losses.gaze_angular_loss(input_dict['head_a'], head_h)
loss = score_head + score_gaze
avg_train_loss += loss
avg_train_score_g += score_gaze
avg_train_score_h += score_head
loss.backward()
optimizer.step()
scheduler.step(step)
step += 1
saver.save_checkpoint(step)