-
Notifications
You must be signed in to change notification settings - Fork 5
/
test.py
192 lines (161 loc) · 7.11 KB
/
test.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
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
import argparse
import random
import torch
from torch import nn, optim
from torch.utils.tensorboard import SummaryWriter
import numpy as np
from src.models.visual_front import Visual_front
from src.models.audio_front import Audio_front
from src.models.memory import Memory
from src.models.temporal_classifier import Temp_classifier
import os
import torch.distributed as dist
from torch.utils.data import DataLoader
from torch.nn import functional as F
from src.data.vid_aud_lrw import MultiDataset
from torch.nn.parallel import DistributedDataParallel as DDP
from torch.nn import DataParallel as DP
import torch.utils.data.distributed
import torch.nn.parallel
import math
import time
import glob
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument('--lrw', default="Data_dir")
parser.add_argument('--model', default="Resnet18")
parser.add_argument("--checkpoint_dir", type=str, default='./data/checkpoints/LRW_18_mstcn_MHVAM')
parser.add_argument("--checkpoint", type=str, default='Checkpoint_dir')
parser.add_argument("--batch_size", type=int, default=60)
parser.add_argument("--lr", type=float, default=0.001)
parser.add_argument("--weight_decay", type=float, default=0.00001)
parser.add_argument("--workers", type=int, default=3)
parser.add_argument("--seed", type=int, default=1)
parser.add_argument("--radius", type=float, default=16.0)
parser.add_argument("--slot", type=int, default=112)
parser.add_argument("--head", type=int, default=8)
parser.add_argument("--max_timesteps", type=int, default=29)
parser.add_argument("--test_aug", default=False, action='store_true')
parser.add_argument("--dataparallel", default=False, action='store_true')
parser.add_argument("--distributed", default=False, action='store_true')
parser.add_argument("--local_rank", type=int, default=0)
parser.add_argument("--gpu", type=str, default='0')
args = parser.parse_args()
return args
def train_net(args):
torch.backends.cudnn.deterministic = False
torch.backends.cudnn.benchmark = True
torch.manual_seed(args.local_rank)
torch.cuda.manual_seed_all(args.local_rank)
random.seed(args.local_rank)
os.environ['OMP_NUM_THREADS'] = '2'
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
os.environ['MASTER_PORT'] = '5555'
if args.distributed:
torch.cuda.set_device(args.local_rank)
torch.distributed.init_process_group(backend='nccl',
init_method='env://')
v_front = Visual_front(in_channels=1)
a_front = Audio_front(in_channels=1)
mem = Memory(radius=args.radius, n_slot=args.slot, n_head=args.head)
tcn = Temp_classifier(radius=args.radius, n_slot=args.slot, head=args.head)
if args.checkpoint is not None:
if args.local_rank == 0:
print(f"Loading checkpoint: {args.checkpoint}")
checkpoint = torch.load(args.checkpoint, map_location=lambda storage, loc: storage.cuda(args.local_rank))
a_front.load_state_dict(checkpoint['a_front_state_dict'])
v_front.load_state_dict(checkpoint['v_front_state_dict'])
mem.load_state_dict(checkpoint['mem_state_dict'])
tcn.load_state_dict(checkpoint['tcn_state_dict'])
del checkpoint
v_front.cuda()
a_front.cuda()
mem.cuda()
tcn.cuda()
if args.distributed:
v_front = torch.nn.SyncBatchNorm.convert_sync_batchnorm(v_front)
a_front = torch.nn.SyncBatchNorm.convert_sync_batchnorm(a_front)
mem = torch.nn.SyncBatchNorm.convert_sync_batchnorm(mem)
tcn = torch.nn.SyncBatchNorm.convert_sync_batchnorm(tcn)
if args.distributed:
v_front = DDP(v_front, device_ids=[args.local_rank], output_device=args.local_rank)
a_front = DDP(a_front, device_ids=[args.local_rank], output_device=args.local_rank)
mem = DDP(mem, device_ids=[args.local_rank], output_device=args.local_rank)
tcn = DDP(tcn, device_ids=[args.local_rank], output_device=args.local_rank)
elif args.dataparallel:
v_front = DP(v_front)
a_front = DP(a_front)
mem = DP(mem)
tcn = DP(tcn)
_ = test(v_front, a_front, mem, tcn)
def test(v_front, a_front, mem, tcn, fast_validate=False):
with torch.no_grad():
a_front.eval()
v_front.eval()
mem.eval()
tcn.eval()
val_data = MultiDataset(
lrw=args.lrw,
mode='test',
max_v_timesteps=args.max_timesteps,
augmentations=False,
)
dataloader = DataLoader(
val_data,
shuffle=False,
batch_size=args.batch_size * 2,
num_workers=args.workers,
drop_last=False
)
criterion = nn.CrossEntropyLoss().cuda()
batch_size = dataloader.batch_size
if fast_validate:
samples = min(2 * batch_size, int(len(dataloader.dataset)))
max_batches = 2
else:
samples = int(len(dataloader.dataset))
max_batches = int(len(dataloader))
val_loss = []
tot_cor, tot_v_cor, tot_a_cor, tot_num = 0, 0, 0, 0
description = 'Check test step' if fast_validate else 'Test'
if args.local_rank == 0:
print(description)
for i, batch in enumerate(dataloader):
if args.local_rank == 0 and i % 10 == 0:
if not fast_validate:
print("******** Validation : %d / %d ********" % ((i + 1) * batch_size, samples))
a_in, v_in, target = batch
v_feat = v_front(v_in.cuda()) #B,S,512
a_feat = a_front(a_in.cuda()) #B,S,512
te_fusion, _, _, _ = mem(v_feat, a_feat, inference=True)
te_m_pred, _, _ = tcn(te_fusion, None, infer=True, mode='te')
ori_pred = te_m_pred.clone().cpu()
################## flip ##################
if args.test_aug:
v_in = v_in.flip(4) # B,C,T,H,W --> B,C,T,H,W
v_feat = v_front(v_in.cuda()) #B,S,512
a_feat = a_front(a_in.cuda()) #B,S,512
te_fusion, _, _, _ = mem(v_feat, a_feat, inference=True)
te_m_pred, _, _ = tcn(te_fusion, None, infer=True, mode='te')
else:
ori_pred = 0.
loss = criterion(te_m_pred, target.long().cuda()).cpu().item()
prediction = torch.argmax(te_m_pred.cpu() + ori_pred, dim=1).numpy()
tot_cor += np.sum(prediction == target.long().numpy())
tot_num += len(prediction)
batch_size = te_m_pred.size(0)
val_loss.append(loss)
if i >= max_batches:
break
a_front.train()
v_front.train()
mem.train()
tcn.train()
print('Test_ACC:', tot_cor / tot_num)
if fast_validate:
return {}
else:
return np.mean(np.array(val_loss)), tot_cor / tot_num
if __name__ == "__main__":
args = parse_args()
train_net(args)