-
Notifications
You must be signed in to change notification settings - Fork 9
/
plot_results.py
172 lines (132 loc) · 5.96 KB
/
plot_results.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
import matplotlib.pyplot as plt
import numpy as np
import queue
import pickle
import os
from datasets.kitti import KITTI
from datasets.utils import euler_to_rotation
def save_trajectory(poses, sequence, save_dir):
"""
Save predicted poses in .txt file
Args:
poses {ndarray}: list with all 4x4 pose matrix
sequence {str}: sequence of KITTI dataset
save_dir {str}: path to save pose
"""
# create directory
if not os.path.exists(save_dir):
os.makedirs(save_dir)
output_filename = os.path.join(save_dir, "{}.txt".format(sequence))
with open(output_filename, "w") as f:
for pose in poses:
pose = pose.flatten()[:12]
line = " ".join([str(x) for x in pose]) + "\n"
f.write(line)
def post_processing(pred_poses, args):
if args["window_size"] == 2:
pred_poses = pred_poses.squeeze(1)
return np.asarray(pred_poses)
num_batchs = pred_poses.shape[0]
# get poses in overlaped frames
q = queue.Queue(args["window_size"]-1) #The max size is 5.
idx = 0
poses = []
while not q.full():
q.put(pred_poses[idx, :, :])
idx = idx + 1
while idx < num_batchs:
# process first full queue
if idx == (args["window_size"]-1):
poses.append(q.queue[0][0, :])
# implemented for specific case window_size = 3 and overlap = 2
avg_pose = (q.queue[0][1, :] + q.queue[1][0, :])/2
poses.append(avg_pose)
if args["window_size"] == 4:
# implemented for specific case window_size = 4 and overlap = 3
avg_pose = (q.queue[0][2, :] + q.queue[1][1, :] + q.queue[2][0, :])/3
poses.append(avg_pose)
elif idx < (num_batchs - 1):
if args["window_size"] == 3:
# implemented for specific case window_size = 3 and overlap = 2
avg_pose = (q.queue[0][1, :] + q.queue[1][0, :])/2
poses.append(avg_pose)
elif args["window_size"] == 4:
# implemented for specific case window_size = 4 and overlap = 3
avg_pose = (q.queue[0][2, :] + q.queue[1][1, :] + q.queue[2][0, :])/3
poses.append(avg_pose)
# process last full queue (idx == num_batchs-1)
else:
if args["window_size"] == 3:
# implemented for specific case window_size = 3 and overlap = 2
poses.append(q.queue[1][1,:])
elif args["window_size"] == 4:
# implemented for specific case window_size = 4 and overlap = 2
avg_pose = (q.queue[1][2, :] + q.queue[2][1, :])/2
poses.append(avg_pose)
poses.append(q.queue[2][2,:])
idx = idx + 1
# update queue
if idx < (num_batchs-1):
idx = idx + 1
first = q.get() #dequeue first element
q.put(pred_poses[idx, :, :])
return np.asarray(poses)
def recover_trajectory_and_poses(poses):
predicted_poses = []
# recover predicted trajectory
predicted_trajectory = []
for i in range(len(poses)-1):
if i == 0:
T = np.eye(4)
angles = poses[i, :3]
t = poses[i, 3:]
# undo normalization
mean_angles = np.array([1.7061e-5, 9.5582e-4, -5.5258e-5])
std_angles = np.array([2.8256e-3, 1.7771e-2, 3.2326e-3])
mean_t = np.array([-8.6736e-5, -1.6038e-2, 9.0033e-1])
std_t = np.array([2.5584e-2, 1.8545e-2, 3.0352e-1])
[x, y, z] = np.multiply(angles, std_angles) + mean_angles
t = np.multiply(t, std_t) + mean_t
R = np.asarray(euler_to_rotation(x, y, z, seq = 'zyx'))
T_r = np.concatenate((np.concatenate([R, np.reshape(t, (3,1))], axis=1) , [[0.0, 0.0, 0.0, 1.0]] ), axis=0)
T_abs = np.dot(T,T_r)
T = T_abs
predicted_poses.append(T)
predicted_trajectory.append(T_abs[:3, 3])
return predicted_poses, predicted_trajectory
if __name__ == "__main__":
ckpt_path = "checkpoints/Exp1/"
ckpt_name = "checkpoint_e80"
sequences = ["01", "03", "04", "05", "06", "07", "10"]
# read hyperparameters and configuration
with open(os.path.join(ckpt_path, "args.pkl"), 'rb') as f:
args = pickle.load(f)
f.close()
ckpt_path = os.path.join(ckpt_path, ckpt_name)
args["checkpoint_path"] = ckpt_path
# plot trajectory and ground truth
for sequence in sequences:
# read ground test data and predicted poses
pred_path = os.path.join(args["checkpoint_path"], "pred_poses_{}.npy".format(sequence))
pred_poses = np.load(pred_path)
# post processing and recover trajectory
poses = post_processing(pred_poses, args)
pred_poses, pred_trajectory = recover_trajectory_and_poses(poses)
save_trajectory(pred_poses, sequence,
save_dir=os.path.join(args["checkpoint_path"], "pred_poses"))
# get ground truth trajectories
test_data = KITTI(sequences=[sequence], window_size=args["window_size"])
gt_poses = test_data.windowed_data.loc[test_data.windowed_data["sequence"]==sequence, [3, 7, 11]]
plt.figure()
pred_trajectory = np.asarray(pred_trajectory)
plt.plot([x[0] for x in pred_trajectory], [z[2] for z in pred_trajectory], "b") # plot estimated trajectory
plt.plot([x[0] for x in gt_poses.values], [z[2] for z in gt_poses.values], "r") # plot ground truth trajectory
plt.grid()
plt.title("VO - Seq {}".format(sequence))
plt.xlabel("Translation in x direction [m]")
plt.ylabel("Translation in z direction [m]")
plt.legend(["estimated", "ground truth"]);
# create checkpoints folder
if not os.path.exists(os.path.join(args["checkpoint_path"], "plots")):
os.makedirs(os.path.join(args["checkpoint_path"], "plots"))
plt.savefig(os.path.join(args["checkpoint_path"], "plots", "pred_traj_{}.png".format(sequence)))