forked from XinJCheng/CSPN
-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
222 lines (198 loc) · 10.4 KB
/
utils.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
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
# -*- coding: utf-8 -*-
"""
Created on Sat Feb 3 16:27:01 2018
@author: norbot
"""
import torch
import math
import os
import data_transform
from torchvision import transforms
from PIL import Image, ImageOps
import matplotlib.pyplot as plt
def max_of_two(y_over_z, z_over_y):
return torch.max(y_over_z, z_over_y)
def evaluate_error(gt_depth, pred_depth):
# for numerical stability
depth_mask = gt_depth>0.0001
batch_size = gt_depth.size(0)
error = {'MSE':0, 'RMSE':0, 'ABS_REL':0, 'LG10':0, 'MAE':0,\
'DELTA1.02':0, 'DELTA1.05':0, 'DELTA1.10':0, \
'DELTA1.25':0, 'DELTA1.25^2':0, 'DELTA1.25^3':0,
}
_pred_depth = pred_depth[depth_mask]
_gt_depth = gt_depth[depth_mask]
n_valid_element = _gt_depth.size(0)
if n_valid_element > 0:
diff_mat = torch.abs(_gt_depth-_pred_depth)
rel_mat = torch.div(diff_mat, _gt_depth)
error['MSE'] = torch.sum(torch.pow(diff_mat, 2))/n_valid_element
error['RMSE'] = math.sqrt(error['MSE'])
error['MAE'] = torch.sum(diff_mat)/n_valid_element
error['ABS_REL'] = torch.sum(rel_mat)/n_valid_element
y_over_z = torch.div(_gt_depth, _pred_depth)
z_over_y = torch.div(_pred_depth, _gt_depth)
max_ratio = max_of_two(y_over_z, z_over_y)
error['DELTA1.02'] = torch.sum(max_ratio < 1.02).numpy()/float(n_valid_element)
error['DELTA1.05'] = torch.sum(max_ratio < 1.05).numpy()/float(n_valid_element)
error['DELTA1.10'] = torch.sum(max_ratio < 1.10).numpy()/float(n_valid_element)
error['DELTA1.25'] = torch.sum(max_ratio < 1.25).numpy()/float(n_valid_element)
error['DELTA1.25^2'] = torch.sum(max_ratio < 1.25**2).numpy()/float(n_valid_element)
error['DELTA1.25^3'] = torch.sum(max_ratio < 1.25**3).numpy()/float(n_valid_element)
return error
# avg the error
def avg_error(error_sum, error_step, total_step, batch_size):
error_avg = {'MSE':0, 'RMSE':0, 'ABS_REL':0, 'LG10':0, 'MAE':0,\
'DELTA1.02':0, 'DELTA1.05':0, 'DELTA1.10':0, \
'DELTA1.25':0, 'DELTA1.25^2':0, 'DELTA1.25^3':0,}
for item, value in error_step.items():
error_sum[item] += error_step[item] * batch_size
error_avg[item] = error_sum[item]/float(total_step)
return error_avg
# print error
def print_error(split, epoch, step, loss, error, error_avg, print_out=False):
format_str = ('%s ===>\n\
Epoch: %d, step: %d, loss=%.4f\n\
MSE=%.4f(%.4f)\tRMSE=%.4f(%.4f)\tMAE=%.4f(%.4f)\tABS_REL=%.4f(%.4f)\n\
DELTA1.02=%.4f(%.4f)\tDELTA1.05=%.4f(%.4f)\tDELTA1.10=%.4f(%.4f)\n\
DELTA1.25=%.4f(%.4f)\tDELTA1.25^2=%.4f(%.4f)\tDELTA1.25^3=%.4f(%.4f)\n')
error_str = format_str % (split, epoch, step, loss,\
error['MSE'], error_avg['MSE'], error['RMSE'], error_avg['RMSE'],\
error['MAE'], error_avg['MAE'], error['ABS_REL'], error_avg['ABS_REL'],\
error['DELTA1.02'], error_avg['DELTA1.02'], \
error['DELTA1.05'], error_avg['DELTA1.05'], \
error['DELTA1.10'], error_avg['DELTA1.10'], \
error['DELTA1.25'], error_avg['DELTA1.25'], \
error['DELTA1.25^2'], error_avg['DELTA1.25^2'], \
error['DELTA1.25^3'], error_avg['DELTA1.25^3'])
if print_out:
print(error_str)
return error_str
def print_single_error(epoch, step, loss, error):
format_str = ('%s ===>\n\
Epoch: %d, step: %d, loss=%.4f\n\
MSE=%.4f\tRMSE=%.4f\tMAE=%.4f\tABS_REL=%.4f\n\
DELTA1.02=%.4f\tDELTA1.05=%.4f\tDELTA1.10=%.4f\n\
DELTA1.25=%.4f\tDELTA1.25^2=%.4f\tDELTA1.25^3=%.4f\n')
print (format_str % ('eval_avg_error', epoch, step, loss,\
error['MSE'], error['RMSE'], error['MAE'], error['ABS_REL'], \
error['DELTA1.02'], error['DELTA1.05'], error['DELTA1.10'], \
error['DELTA1.25'], error['DELTA1.25^2'], error['DELTA1.25^3']))
# update_best_model
def updata_best_model(error_avg, best_RMSE):
if error_avg['RMSE'] < best_RMSE:
return True
else:
return False
# log best_model
def log_file_folder_make(save_dir):
if not os.path.isdir(save_dir):
os.makedirs(save_dir, 0o777)
train_log_file = os.path.join(save_dir, 'log_train.txt')
train_fd = open(train_log_file, 'w')
train_fd.write('epoch\t bestModel\t MSE\t RMSE\t MAE\t \
DELTA1.02\t DELTA1.05\t DELTA1.10\t DELTA1.25\t \
DELTA1.25^2\t DELTA1.25^3\t ABS_REL\n')
train_fd.close()
eval_log_file = os.path.join(save_dir, 'log_eval.txt')
eval_fd = open(eval_log_file, 'w')
eval_fd.write('epoch\t bestModel\t MSE\t RMSE\t MAE\t \
DELTA1.02\t DELTA1.05\t DELTA1.10\t \
DELTA1.25\t DELTA1.25^2\t DELTA1.25^3\t ABS_REL\n')
eval_fd.close()
def log_result(save_dir, error_avg, epoch, lr, best_model, split):
format_str = ('%.4f\t %.4f\t\t %.4f\t %.4f\t %.4f\t %.4f\t %.4f\t %.4f\t %.4f\t %.4f\t %.4f\t %.4f\n')
train_log_file = os.path.join(save_dir, 'log_train.txt')
eval_log_file = os.path.join(save_dir, 'log_eval.txt')
if split == 'train':
train_fd = open(train_log_file, 'a')
train_fd.write(format_str%(epoch, best_model, error_avg['MSE'], error_avg['RMSE'],\
error_avg['MAE'], error_avg['DELTA1.02'], error_avg['DELTA1.05'],\
error_avg['DELTA1.10'], error_avg['DELTA1.25'], error_avg['DELTA1.25^2'],\
error_avg['DELTA1.25^3'], error_avg['ABS_REL']))
train_fd.close()
elif split == 'eval':
eval_fd = open(eval_log_file, 'a')
eval_fd.write(format_str%(epoch, best_model, error_avg['MSE'], error_avg['RMSE'],\
error_avg['MAE'], error_avg['DELTA1.02'], error_avg['DELTA1.05'],\
error_avg['DELTA1.10'], error_avg['DELTA1.25'], error_avg['DELTA1.25^2'],\
error_avg['DELTA1.25^3'], error_avg['ABS_REL']))
eval_fd.close()
# log best_model
def log_file_folder_make_lr(save_dir):
if not os.path.isdir(save_dir):
os.makedirs(save_dir, 0o777)
train_log_file = os.path.join(save_dir, 'log_train.txt')
train_fd = open(train_log_file, 'w')
train_fd.write('epoch\t lr\t bestModel\t MSE\t RMSE\t MAE\t \
DELTA1.02\t DELTA1.05\t DELTA1.10\t DELTA1.25\t \
DELTA1.25^2\t DELTA1.25^3\t ABS_REL\n')
train_fd.close()
eval_log_file = os.path.join(save_dir, 'log_eval.txt')
eval_fd = open(eval_log_file, 'w')
eval_fd.write('epoch\t lr\t bestModel\t MSE\t RMSE\t MAE\t \
DELTA1.02\t DELTA1.05\t DELTA1.10\t DELTA1.25\t \
DELTA1.25^2\t DELTA1.25^3\t ABS_REL\n')
eval_fd.close()
def log_result_lr(save_dir, error_avg, epoch, lr, best_model, split):
format_str = ('%.4f\t %.4f\t %.4f\t\t %.4f\t %.4f\t %.4f\t %.4f\t %.4f\t %.4f\t %.4f\t %.4f\t %.4f\t %.4f\n')
train_log_file = os.path.join(save_dir, 'log_train.txt')
eval_log_file = os.path.join(save_dir, 'log_eval.txt')
if split == 'train':
train_fd = open(train_log_file, 'a')
train_fd.write(format_str%(epoch, lr, best_model, error_avg['MSE'], error_avg['RMSE'],\
error_avg['MAE'], error_avg['DELTA1.02'], error_avg['DELTA1.05'],\
error_avg['DELTA1.10'], error_avg['DELTA1.25'], error_avg['DELTA1.25^2'],\
error_avg['DELTA1.25^3'], error_avg['ABS_REL']))
train_fd.close()
elif split == 'eval':
eval_fd = open(eval_log_file, 'a')
eval_fd.write(format_str%(epoch, lr, best_model, error_avg['MSE'], error_avg['RMSE'],\
error_avg['MAE'], error_avg['DELTA1.02'], error_avg['DELTA1.05'],\
error_avg['DELTA1.10'], error_avg['DELTA1.25'], error_avg['DELTA1.25^2'],\
error_avg['DELTA1.25^3'], error_avg['ABS_REL']))
eval_fd.close()
def un_normalize(tensor):
img_mean = (0.485, 0.456, 0.406)
img_std = (0.229, 0.224, 0.225)
for t, m, s in zip(tensor, img_mean, img_std):
t.mul_(s).add_(m)
return tensor
def save_eval_img(data_set, model_dir, index, input_rgbd, input_rgb, gt_depth, pred_depth):
img_save_folder = os.path.join(model_dir, 'eval_result')
if not os.path.isdir(img_save_folder):
os.makedirs(img_save_folder, 0o777)
save_name_rgb = os.path.join(img_save_folder, "%05d_input.png" % (index))
save_name_gt = os.path.join(img_save_folder, "%05d_gt.png" % (index))
save_name_pred = os.path.join(img_save_folder, "%05d_pred.png" % (index))
save_name_sparse_point = os.path.join(img_save_folder, "%05d_sparse_point.png" % (index))
save_name_sparse_mask = os.path.join(img_save_folder, "%05d_sparse_mask.png" % (index))
save_rgb = transforms.ToPILImage()(torch.squeeze(input_rgb, 0))
save_gt = None
save_pred = None
if data_set == 'kitti':
save_sparse_point = data_transform.ToPILImage()(input_rgbd[:,3,:,:])
save_sparse_mask = data_transform.ToPILImage()(input_rgbd[:,3,:,:].sign())
save_gt = data_transform.ToPILImage()(torch.squeeze(gt_depth*1.0, 0))
save_pred = data_transform.ToPILImage()(torch.squeeze(pred_depth*1.0, 0))
# print("save_rgb shape: {}".format(input_rgb.shape))
save_rgb.save(save_name_rgb)
save_gt.save(save_name_gt)
save_pred.save(save_name_pred)
save_sparse_point.save(save_name_sparse_point)
save_sparse_mask.save(save_name_sparse_mask)
# plt.imsave(save_name_rgb, save_rgb)
# plt.imsave(save_name_gt, save_gt)
# plt.imsave(save_name_pred, save_pred)
elif data_set == 'nyudepth':
save_gt = data_transform.ToPILImage()(torch.squeeze(gt_depth*25.5, 0))
save_pred = data_transform.ToPILImage()(torch.squeeze(pred_depth*25.5, 0))
save_rgb.save(save_name_rgb)
save_gt.save(save_name_gt)
save_pred.save(save_name_pred)
def test_eval_error():
gt_depth = torch.abs(torch.randn(1,3,4))
pred_depth = torch.abs(torch.randn(1,3,4))
eval_result = evaluate_error(gt_depth, pred_depth)
for item, value in eval_result.items():
print(('%s\' value is: %f') %(item, eval_result[item]))