-
Notifications
You must be signed in to change notification settings - Fork 0
/
mdvo_test_pose.py
223 lines (197 loc) · 9.94 KB
/
mdvo_test_pose.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
223
from __future__ import division
import os
import math
import scipy.misc as sm
import tensorflow as tf
import numpy as np
from glob import glob
from mdvo_model import *
from data_loader import DataLoader
from kitti_eval.pose_evaluation_utils import *
from offical_test import *
from kitti_eval.eval_pose import eval_pose
def test_pose(opt):
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3"
if not os.path.isdir(opt.output_dir):
os.makedirs(opt.output_dir)
input_uint8 = tf.placeholder(tf.uint8, [opt.batch_size,
opt.img_height, opt.img_width, opt.seq_length * 3],
name='raw_input')
input_uint8_r = tf.placeholder(tf.uint8, [opt.batch_size,
opt.img_height, opt.img_width, opt.seq_length * 3],
name='raw_input_r')
tgt_image = input_uint8[:, :, :, :3]
tgt_image_r = input_uint8_r[:, :, :, :3]
src_image_stack = input_uint8[:, :, :, 3:]
src_image_stack_r = input_uint8_r[:, :, :, 3:]
intrinsics = tf.placeholder(tf.float32, [opt.batch_size, 3, 3], name='intrinsics_input')
loader = DataLoader(opt)
intrinsics_ms = loader.get_multi_scale_intrinsics(intrinsics, opt.num_scales)
model = RA_GANVO_Model(opt, tgt_image, src_image_stack, intrinsics_ms, tgt_image_r, src_image_stack_r)
fetches = {"pose": model.pred_rt_full}
saver = tf.train.Saver([var for var in tf.model_variables()])
# #### load test frames #####
seq_dir = os.path.join(opt.dataset_dir, 'sequences', '%.2d' % opt.pose_test_seq)
img_dir = os.path.join(seq_dir, 'image_2')
N = len(glob(img_dir + '/*.png'))
test_frames = ['%.2d %.6d' % (opt.pose_test_seq, n) for n in range(N)]
# #### load time file #####
with open(opt.dataset_dir + '/sequences/%.2d/times.txt' % opt.pose_test_seq, 'r') as f:
times = f.readlines()
times = np.array([float(s[:-1]) for s in times])
# #### Go! #####
max_src_offset = (opt.seq_length - 1)
all_pose = []
pose_compare = []
pred_all_list = []
# ################
# ## need argus ##
# ################
out_file = opt.output_dir + '/tre_result/'
# one model file load
out_file_model = out_file + opt.init_ckpt_file.split('/')[-1][6:] + '/'
if not os.path.exists(out_file_model):
os.makedirs(out_file_model)
# txt file load
out_txt_file = out_file_model + '{:02}'.format(opt.pose_test_seq) + '.txt'
# summary result load
result_sum = out_file + 'result_sum.txt'
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.1)
with tf.Session(config=tf.ConfigProto(gpu_options=gpu_options)) as sess:
saver.restore(sess, opt.init_ckpt_file)
final_pose = None
final = N - max_src_offset
with open(out_txt_file, 'w') as f:
for tgt_idx in range(0, N-max_src_offset, opt.batch_size):
if (tgt_idx-max_src_offset) % 100 == 0:
print('Progress: %d/%d' % (tgt_idx-max_src_offset, N))
inputs = np.zeros((opt.batch_size, opt.img_height, opt.img_width, 3*opt.seq_length), dtype=np.uint8)
inputs_r = np.zeros((opt.batch_size, opt.img_height, opt.img_width, 3 * opt.seq_length), dtype=np.uint8)
for b in range(opt.batch_size):
idx = tgt_idx + b
if idx >= N-max_src_offset:
break
image_seq = load_image_sequence(opt.dataset_dir,
test_frames,
idx,
opt.seq_length,
opt.img_height,
opt.img_width)
inputs[b] = image_seq
# add right images of tgt and src
image_seq_r = load_image_sequence_r(opt.dataset_dir,
test_frames,
idx,
opt.seq_length,
opt.img_height,
opt.img_width)
inputs_r[b] = image_seq_r
# get K (intrinsics of camera)
intrinsics_09 = np.array([239.926529, 0., 204.229309, 0., 244.615326, 63.346298, 0., 0., 1.]).reshape([1, 3, 3])
pred = sess.run(fetches, feed_dict={input_uint8: inputs,
input_uint8_r: inputs_r,
intrinsics: intrinsics_09})
pred_poses = pred['pose']
# scale back to absolute scale
for i in range(1, opt.num_source+1):
pred_poses[i][0][0:3, 3] = pred_poses[i][0][0:3, 3] * 0.3087
# get 5frames reslut
pred_all_list.append(pred_poses)
temp_len = len(pred_all_list)
if temp_len >= 4:
for b in range(opt.batch_size):
idx = tgt_idx + b
if idx >= N - max_src_offset:
break
pred_poses_0 = pred_all_list[temp_len - 4] # [0,1-->0]
pred_poses_1 = np.matmul(pred_all_list[temp_len - 4][1], pred_all_list[temp_len - 3][1]) # [2-->0]
pred_poses_2 = np.matmul(pred_poses_1, pred_all_list[temp_len - 2][1]) # [3 --> 0]
pred_poses_3 = np.matmul(pred_poses_2, pred_all_list[temp_len - 1][1]) # [4-->0]
# pred_pose = pred_poses[b]
pred_poses_0.append(pred_poses_1)
pred_poses_0.append(pred_poses_2)
pred_poses_0.append(pred_poses_3) # [0,1->0,2->0,3->0,4->0] (5,1,4,4)
curr_times = times[idx - 2 - max_src_offset:idx - 2 + max_src_offset + 3]
# ate result for len 5
out_file_ate = opt.output_dir + '/ate_result/'
# one model file load
out_file_model_ate = out_file_ate + opt.init_ckpt_file.split('/')[-1][6:] + '/{:02}'.format(opt.pose_test_seq)
# txt file load
if not os.path.exists(out_file_model_ate):
os.makedirs(out_file_model_ate)
out_file = out_file_model_ate + '/%.6d.txt' % (idx - max_src_offset - 2)
dump_pose_seq_TUM(out_file, pred_poses_0, curr_times)
# get tre
for b in range(opt.batch_size):
pred_pose = pred_poses
pred_pose = [np.squeeze(pose) for pose in pred_pose]
pose_compare.append(pred_pose[opt.seq_length - 1])
if final_pose is None:
for i in range(2):
this_pose = pred_pose[i]
out_pose = np.reshape(this_pose, [-1])
out_pose = tuple([out_pose[i] for i in range(12)])
f.write('%f %f %f %f %f %f %f %f %f %f %f %f\n' % out_pose)
all_pose.append(this_pose)
final_pose = pred_pose[1]
else:
this_pose = np.dot(final_pose, pred_pose[1])
final_pose = np.dot(final_pose, pred_pose[1])
out_pose = np.reshape(this_pose, [-1])
out_pose = tuple([out_pose[i] for i in range(12)])
f.write('%f %f %f %f %f %f %f %f %f %f %f %f\n' % out_pose)
all_pose.append(this_pose)
# eval 5frames ATE
pred_dir = out_file_model_ate
gtruth_dir = 'data/pose_gt/ate/{:02}/'.format(opt.pose_test_seq)
test_model = opt.init_ckpt_file.split('/')[-1][6:]
ate_summary_dir = out_file_ate
eval_pose(pred_dir, gtruth_dir, test_model, ate_summary_dir)
# eval tre (error of T and R)
eval_ate = kittiEvalOdom('data/pose_gt/tre/', [opt.pose_test_seq])
sequence = opt.pose_test_seq
model_name = opt.init_ckpt_file.split('/')[-1]
model_index = int(model_name.split('-')[-1])
eval_ate.eval_sum(out_file_model, result_sum, model_index, sequence)
def load_image_sequence(dataset_dir,
frames,
tgt_idx,
seq_length,
img_height,
img_width):
half_offset = int((seq_length - 1))
for o in range(0, half_offset+1):
curr_idx = tgt_idx + o
curr_drive, curr_frame_id = frames[curr_idx].split(' ')
img_file = os.path.join(
dataset_dir, 'sequences', '%s/image_2/%s.png' % (curr_drive, curr_frame_id))
curr_img = sm.imread(img_file)
curr_img = sm.imresize(curr_img, (img_height, img_width))
if o == 0:
image_seq = curr_img
elif o == 1:
image_seq = np.dstack((image_seq, curr_img))
# else:
# image_seq = np.dstack((image_seq, curr_img))
return image_seq
def load_image_sequence_r(dataset_dir,
frames,
tgt_idx,
seq_length,
img_height,
img_width):
half_offset = int((seq_length - 1))
for o in range(0, half_offset+1):
curr_idx = tgt_idx + o
curr_drive, curr_frame_id = frames[curr_idx].split(' ')
img_file = os.path.join(
dataset_dir, 'sequences', '%s/image_3/%s.png' % (curr_drive, curr_frame_id))
curr_img = sm.imread(img_file)
curr_img = sm.imresize(curr_img, (img_height, img_width))
if o == 0:
image_seq = curr_img
elif o == 1:
image_seq = np.dstack((image_seq, curr_img))
# else:
# image_seq = np.dstack((image_seq, curr_img))
return image_seq