-
Notifications
You must be signed in to change notification settings - Fork 2
/
testfmnet_nyu.py
199 lines (141 loc) · 5.79 KB
/
testfmnet_nyu.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
193
194
195
196
197
198
199
import os
os.environ['CUDA_VISIBLE_DEVICES'] = '0,1'
import argparse
import cv2
from os.path import join as pjoin
import json
import math
import numpy as np
from torch.utils.data import Dataset, DataLoader
import torch
from PIL import Image
import matplotlib.pyplot as plt
import random
import os
from torchvision import transforms, utils
import scipy.io as io
import resnext101_wsl
from DepthNet4 import *
from ConvTrans import *
from DepthNet4 import Decoder
from DepthNet4 import Decoder_noAO
import matplotlib.pyplot as plt
from torch.nn import DataParallel
import h5py
import glob
torch.backends.cudnn.enabled = True
torch.backends.cudnn.benchmark = True
def img_loader(path):
image = cv2.imread(path)
if image.ndim == 2:
image = cv2.cvtColor(image, cv2.COLOR_GRAY2BGR)
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) / 255.0
return image
def gt_png_loader(path):
depth = cv2.imread(path, -1)
depth = depth / 1000.0
#real_depth[real_depth!=0] = 1.0 / real_depth[real_depth!=0]
#disp_norm = real_depth / np.max(real_depth)
#valid_mask = real_depth != 0
return depth.astype(np.float32)
def compute_errors(pred, gt):
#pred = torch.clamp(pred, 1e-6, 9.9955)
pred = pred[gt>0]
gt = gt[gt>0]
pred = pred.cpu().numpy()
gt = gt.cpu().numpy()
thresh = np.maximum((gt / pred), (pred / gt))
d1 = (thresh < 1.25).mean()
d2 = (thresh < 1.25 ** 2).mean()
d3 = (thresh < 1.25 ** 3).mean()
rmse = (gt - pred) ** 2
rmse = np.sqrt(rmse.mean())
rmse_log = (np.log(gt) - np.log(pred)) ** 2
rmse_log = np.sqrt(rmse_log.mean())
abs_rel = np.mean(np.abs(gt - pred) / gt)
sq_rel = np.mean(((gt - pred) ** 2) / gt)
err = np.log(pred) - np.log(gt)
silog = np.sqrt(np.mean(err ** 2) - 0.85*np.mean(err) ** 2) * 10
err = np.abs(np.log10(pred) - np.log10(gt))
log10 = np.mean(err)
return rmse, rmse_log, silog, log10, abs_rel, sq_rel, d1, d2, d3
mean = [0.485, 0.456, 0.406]
std = [0.229, 0.224, 0.225]
seq_len= 12
small_seq_len = 2
device_cnn_encoder = torch.device("cuda:0")
device_fmnet_encoder = torch.device("cuda:0")
device_fmnet_decoder = torch.device("cuda:1")
device_cnn_decoder = torch.device("cuda:1")
cnn_encoder = resnext101_wsl.resnext101_wsl() #resnet.resnet50() #resnext101_wsl.resnext101_wsl()
fmnet_Encoder = fmnet_encoder(num_hidden = [2048],batch_size = 1 ,img_width = 20,img_height = 15 ,input_length = seq_len ,small_length = small_seq_len ,encoder_depth = 6, device = device_fmnet_encoder)
fmnet_Decoder = fmnet_decoder(num_hidden = [2048],batch_size = 1 ,img_width = 20,img_height = 15 ,input_length = seq_len ,small_length = small_seq_len ,encoder_depth = 1, device = device_fmnet_decoder)
cnn_decoder = Decoder(inchannels=[256, 512, 1024, 2048], midchannels=[256, 256, 256, 512], upfactors=[2, 2, 2, 2], outchannels=1)
checkpoint = torch.load('./checkpoint/nyu_epoch_20.pth',map_location = 'cpu') #18
cnn_encoder.load_state_dict(checkpoint['cnn_encoder'])
cnn_decoder.load_state_dict(checkpoint['cnn_decoder'])
fmnet_Encoder.load_state_dict(checkpoint['Mae_Encoder'])
fmnet_Decoder.load_state_dict(checkpoint['Mae_Decoder'])
cnn_encoder.to(device_cnn_encoder)
fmnet_Encoder.to(device_fmnet_encoder)
fmnet_Decoder.to(device_fmnet_decoder)
cnn_decoder.to(device_cnn_decoder)
base_dir = './data/testnyu_data/'
alld1,alld2,alld3 = 0,0,0
allrmse, allrmse_log, allsilog, alllog10, allabs_rel, allsq_rel = 0,0,0,0,0,0
for i in range(654):
print('Testing sequence:',i+1,'/654')
image_dir = base_dir + str(i+1) +'/rgb/'
gt_dir = base_dir + str(i+1) +'/gt/'
image_seq = glob.glob(image_dir+'*.png')
image_seq.sort(key=lambda x: int(x[-10:-4]))
gt_pic = glob.glob(gt_dir+'*.png')[0]
pic_num = int(gt_pic[-10:-4])
for j in range(seq_len):
img = img_loader(image_seq[j])
img = (img - mean) / std
img = np.transpose(img, (2, 0, 1))
img = torch.Tensor(np.ascontiguousarray(img).astype(np.float32))
img = img.unsqueeze(0)
if j==0:
img_seq = img
else:
img_seq=torch.cat([img_seq,img],dim=0)
gt = gt_png_loader(gt_pic)
gt = torch.Tensor(np.ascontiguousarray(gt.astype(np.float32)))
with torch.no_grad():
img_seq = img_seq.to(device_cnn_encoder)
img_seq = img_seq.squeeze(0)
spatial_feat0, spatial_feat1, spatial_feat2,outputs = cnn_encoder(img_seq)
outputs = outputs.to(device_fmnet_encoder)
outputs, choose_frames = fmnet_Encoder(outputs)
outputs = outputs.to(device_fmnet_decoder)
outputsall = fmnet_Decoder(outputs, choose_frames)
spatial_feat0 = spatial_feat0.to(device_cnn_decoder)
spatial_feat1 = spatial_feat1.to(device_cnn_decoder)
spatial_feat2 = spatial_feat2.to(device_cnn_decoder)
outputsall = outputsall.to(device_cnn_decoder)
outputsall = cnn_decoder([spatial_feat0, spatial_feat1, spatial_feat2, outputsall])[0]
outputsall = F.relu(outputsall)
pred = outputsall[pic_num].squeeze()
pred = torch.clamp(pred, min=0.7133, max=9.9955)
gt = gt.squeeze()
gt = gt[44:471, 40:601]
pred = pred[44:471, 40:601]
rmse, rmse_log, silog, log10, abs_rel, sq_rel, d1, d2, d3 = compute_errors(pred,gt)
print(d1,d2,d3)
alld1 = alld1 + d1
alld2 = alld2 + d2
alld3 = alld3 + d3
allrmse = allrmse + rmse
allrmse_log = allrmse_log + rmse_log
allsilog = allsilog + silog
alllog10 = alllog10 + log10
allabs_rel = allabs_rel + abs_rel
allsq_rel = allsq_rel + sq_rel
print('d1:',alld1/654)
print('d2:',alld2/654)
print('d3:',alld3/654)
print('RMSE:',allrmse/654)
print('log10',alllog10/654)
print('Absrel:',allabs_rel/654)