forked from donggong1/memae-anomaly-detection
-
Notifications
You must be signed in to change notification settings - Fork 0
/
script_testing.py
168 lines (152 loc) · 5.87 KB
/
script_testing.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
from __future__ import absolute_import, print_function
import os
import utils
import torch
from torchvision import transforms
from torch.utils.data import DataLoader
import numpy as np
import data
import scipy.io as sio
from options.testing_options import TestOptions
import utils
import time
from tqdm import tqdm
from models import AutoEncoderCov3D, AutoEncoderCov3DMem
###
opt_parser = TestOptions()
opt = opt_parser.parse(is_print=True)
use_cuda = opt.UseCUDA
#device = torch.device('cpu')
device = torch.device("cuda" if use_cuda else "cpu")
print(f'using {device} now')
###
batch_size_in = opt.BatchSize # 1
chnum_in_ = opt.ImgChnNum # channel number of the input images
framenum_in_ = opt.FrameNum # frame number of the input images in a video clip
mem_dim_in = opt.MemDim
sparse_shrink_thres = opt.ShrinkThres
img_crop_size = 0
######
model_setting = utils.get_model_setting(opt)
# data path
data_root = opt.DataRoot + opt.Dataset + '/'
data_frame_dir = data_root + 'Test/'
data_idx_dir = data_root + 'Test_idx/'
print(f'data_root: {data_root}')
# model path
model_root = opt.ModelRoot
if(opt.ModelFilePath):
model_path = opt.ModelFilePath
else:
model_path = os.path.join(model_root, model_setting + '.pt')
print(f'model path: {model_path}')
# test result path
te_res_root = opt.OutRoot
te_res_path = te_res_root + '/' + 'res_' + model_setting
utils.mkdir(te_res_path)
# loading trained model
if (opt.ModelName == 'AE'):
model = AutoEncoderCov3D(chnum_in_)
elif(opt.ModelName == 'MemAE'):
model = AutoEncoderCov3DMem(
chnum_in_, mem_dim_in, shrink_thres=sparse_shrink_thres)
else:
model = []
print('Wrong Name.')
# print(f'model:{model}')
##
# torch.backends.cudnn.enabled = False
model_para = torch.load(model_path, map_location=torch.device('cpu'))
model.load_state_dict(model_para)
print('load complete')
model.to(device)
model.eval()
print('start to work')
##
if(chnum_in_ == 1):
norm_mean = [0.5]
norm_std = [0.5]
elif(chnum_in_ == 3):
norm_mean = (0.5, 0.5, 0.5)
norm_std = (0.5, 0.5, 0.5)
frame_trans = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(norm_mean, norm_std)
])
unorm_trans = utils.UnNormalize(mean=norm_mean, std=norm_std)
# ##
video_list = utils.get_subdir_list(data_idx_dir)
video_num = len(video_list)
##
with torch.no_grad():
for ite_vid in tqdm(range(video_num)):
video_name = video_list[ite_vid]
# idx path of the current sub dir
video_idx_path = os.path.join(data_idx_dir, video_name)
# frame path of the current sub dir
video_frame_path = os.path.join(data_frame_dir, video_name)
# info for current video
idx_name_list = [name for name in os.listdir(video_idx_path)
if os.path.isfile(os.path.join(video_idx_path, name))]
idx_name_list.sort()
# load data (frame clips) for single video
video_dataset = data.VideoDatasetOneDir(
video_idx_path, video_frame_path, transform=frame_trans)
video_data_loader = DataLoader(video_dataset,
batch_size=batch_size_in,
shuffle=False
)
# testing results on a single video sequence
print('[vidx %02d/%d] [vname %s]' % (ite_vid+1, video_num, video_name))
recon_error_list = []
#
for batch_idx, (item, frames) in tqdm(enumerate(video_data_loader)):
idx_name = idx_name_list[item[0]]
idx_data = sio.loadmat(os.path.join(video_idx_path, idx_name))
v_name = idx_data['v_name'][0] # video name
# frame index list for a video clip
frame_idx = idx_data['idx'][0, :]
######
frames = frames.to(device)
##
if (opt.ModelName == 'AE'):
recon_frames = model(frames)
# calculate reconstruction error (MSE)
recon_np = utils.vframes2imgs(unorm_trans(
recon_frames.data), step=1, batch_idx=0)
input_np = utils.vframes2imgs(
unorm_trans(frames.data), step=1, batch_idx=0)
r = utils.crop_image(recon_np, img_crop_size) - \
utils.crop_image(input_np, img_crop_size)
# recon_error = np.mean(sum(r**2)**0.5)
recon_error = np.mean(r ** 2) # **0.5
recon_error_list += [recon_error]
elif (opt.ModelName == 'MemAE'):
recon_res = model(frames)
recon_frames = recon_res['output']
r = recon_frames - frames
r = utils.crop_image(r, img_crop_size)
sp_error_map = torch.sum(r**2, dim=1)**0.5
s = sp_error_map.size()
sp_error_vec = sp_error_map.view(s[0], -1)
recon_error = torch.mean(sp_error_vec, dim=-1)
recon_error_list += recon_error.cpu().tolist()
######
# elif (opt.ModelName == 'MemAE'):
# recon_res = model(frames)
# recon_frames = recon_res['output']
# recon_np = utils.btv2btf(unorm_trans(recon_frames.data))
# input_np = utils.btv2btf(unorm_trans(frames.data))
# r = utils.crop_image(recon_np, img_crop_size) - utils.crop_image(input_np, img_crop_size)
# sp_error_map = np.sum(r**2, axis=1)**0.5
# tmp_s = sp_error_map.shape
# sp_error_vec = np.reshape(sp_error_map, (tmp_s[0], -1))
# recon_error = np.mean(sp_error_vec, axis=-1)
# recon_error_list += recon_error.tolist()
#######
else:
recon_error = -1
print('Wrong ModelName.')
np.save(os.path.join(te_res_path, video_name + '.npy'), recon_error_list)
# evaluation
utils.eval_video(data_root, te_res_path, is_show=False)