-
Notifications
You must be signed in to change notification settings - Fork 0
/
hand_eye_infer.py
195 lines (156 loc) · 7.6 KB
/
hand_eye_infer.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
import torch
import numpy as np
import time
import pandas as pd
import os
import argparse
import open3d as o3d
from utils.Handeye_datasets import Handeye_datasets
import utils.config as con
def infer_vis():
model = torch.load(con.model_path)#trained model path and loading
#load data
sampled_profilesLabel = pd.read_csv(con.vis_file_path.replace('\n', ''), sep=',', header=None)
sampled_profilesLabel = np.array(sampled_profilesLabel)[0:con.points_n, :]
#fill points
if sampled_profilesLabel.shape[0] < con.points_n:
sub_array = np.array([-199.0, -1, 0]) if sampled_profilesLabel.shape[1] == 3 else np.array([-199.0, -1])
sub_array = np.repeat(sub_array.reshape(1, -1), con.points_n - sampled_profilesLabel.shape[0], axis=0)
sampled_profilesLabel = np.vstack([sampled_profilesLabel, sub_array])
sampled_profilesLabel = torch.tensor(sampled_profilesLabel, dtype=torch.float32).unsqueeze(0)
un_profile_r = sampled_profilesLabel[:, :, 0:2]
un_profile = un_profile_r[:, :, [0, 1]]
#inference
torch.cuda.synchronize()
forward_start_time = time.time()
un_profile = un_profile.cuda()
centroid = torch.mean(un_profile, axis=1)
un_profile = un_profile - centroid.unsqueeze(1)#these 2 steps consume a lot of time
norm_unprofile = torch.max(torch.sqrt(torch.sum(un_profile ** 2, axis=2)), axis=1, keepdim=True)[0]
un_profile = un_profile / norm_unprofile.unsqueeze(1)
with torch.no_grad():
model.eval()
un_profile = un_profile.transpose(2, 1)
model = model.cuda()
out_l = model(un_profile)
_, pred = torch.max(out_l, 1)
torch.cuda.synchronize()
forward_time_cost_100 = (time.time() - forward_start_time)
print('Forward time:%.4f' % forward_time_cost_100)
#plot and visualization
points_p = un_profile_r.view(-1, 2).numpy()
points_p = np.insert(points_p, 1, 0, axis=1)#add y axis to plot
label_p = pred.cpu().view(-1, 1).numpy()
color = np.zeros((points_p.shape[0], 3))
for j in range(con.points_n):
if label_p[j, 0] == 0:
color[j, :] = [1, 1, 0]#[1, 1, 0]
elif label_p[j, 0] == 1:
color[j, :] = [0, 0, 1]
elif label_p[j, 0] == 2:
color[j, :] = [0, 1, 0]
elif label_p[j, 0] == 3:
color[j, :] = [1, 0, 0]#1,0,0
elif label_p[j, 0] == 4:
color[j, :] = [127 / 255, 0, 55 / 255] # 0,0,1
elif label_p[j, 0] == 5:
color[j, :] = [205/255, 106/255, 0]#0,0,1
point_cloud = o3d.geometry.PointCloud()
point_cloud.points = o3d.utility.Vector3dVector(points_p)
point_cloud.colors = o3d.utility.Vector3dVector(color)
o3d.visualization.draw_geometries_with_editing([point_cloud])
pd.DataFrame(np.hstack([points_p, label_p])).to_csv('./test.csv', header=0, index=0, float_format='%.3f', sep=',')#save the visulization results
def infer_results():
num_classes = 2#segmentation classes number
model = torch.load(con.model_path) # trained model path and loading
test_path = os.path.join(con.dataset_pre, con.test_file_name)
seg_classes = {0: 'Outlier', 1: 'Circle points'}
shape_ious = {'Outlier': [], 'Circle points': []}
shape_ious_value = []
total_correctv = 0
total_points = 0
total_seen_class = [0 for _ in range(len(seg_classes))]
total_correct_class = [0 for _ in range(len(seg_classes))]
test_dataset = Handeye_datasets(test_path, con.dataset_pre+'test_set', normal_method='Whole', num_classes=num_classes, dim=2)
test_dataloader = torch.utils.data.DataLoader(
test_dataset,
batch_size=con.infer_batch_size,
shuffle=True,
num_workers=4,
pin_memory=False
)
with torch.no_grad():
model = model.eval()
for batch_i, (profile_imv, profile_labelv, _, _) in enumerate(test_dataloader):
profile_imv = profile_imv.float().permute(0, 2, 1)
profile_labelv = profile_labelv.long().squeeze(2)
if torch.cuda.is_available():
profile_imv = torch.autograd.Variable(profile_imv).cuda()
profile_labelv = torch.autograd.Variable(profile_labelv).cuda()
out_v = model(profile_imv)
_, pred_v = torch.max(out_v, 1)
pred_np = pred_v.long().cpu().numpy()
label_np = profile_labelv.long().cpu().numpy()
num_correctv = np.sum(pred_np == label_np)
total_correctv += num_correctv
total_points += pred_v.cpu().numpy().shape[0] * pred_v.cpu().numpy().shape[1]
for j in range(len(seg_classes)):
if np.sum((pred_np == j) | (label_np == j)) != 0:
total_correct_class[j] += np.sum((pred_np == j) & (label_np == j))
total_seen_class[j] += np.sum((pred_np == j) | (label_np == j))
for j in range(len(seg_classes)):
if total_seen_class[j] != 0:
iou = total_correct_class[j] / float(total_seen_class[j])
shape_ious_value.append(iou)
shape_ious[seg_classes[j]] = iou
else:
shape_ious[seg_classes[j]] = 'None'
for key, value in shape_ious.items():
if type(value) != str:
print('%s IOU:%.5f' % (key, value))
else:
print('%s IOU:%S' % (key, value))
print('Mean IOU: %.5f' % np.mean(shape_ious_value))
print('val_acc:%.5f' % (total_correctv / float(total_points)))
def batch_circle_segmentation():
num_classes = 2
if not os.path.exists(con.output_base_dir):
os.mkdir(con.output_base_dir)
model = torch.load(con.model_path) # trained model path and loading
val_path = os.path.join(con.dataset_pre, con.test_file_name)
val_dataset = Handeye_datasets(val_path, con.dataset_pre+'test_set', normal_method='Whole', num_classes=num_classes, dim=2)
val_dataloader = torch.utils.data.DataLoader(
val_dataset,
batch_size=1,
shuffle=True,
num_workers=4,
pin_memory=False
)
with torch.no_grad():
for batch_i, (profile_imv, profile_labelv, file_path, point_on) in enumerate(val_dataloader):
profile_imv = profile_imv.float().permute(0, 2, 1)
if torch.cuda.is_available():
profile_imv = torch.autograd.Variable(profile_imv).cuda()
out_v = model(profile_imv)
_, pred_v = torch.max(out_v, 1)
pred_np = pred_v.long().permute(1, 0).cpu().numpy()
profile_st = pd.read_csv(os.path.join(con.dataset_pre + 'test_set', file_path[0][1:]), sep=',', header=None)
profile_st = np.array(profile_st).astype('float')
out_p_l = np.hstack([profile_st[:, 0:2], pred_np[:point_on, :]])
save_name = os.path.basename(file_path[0])
pd.DataFrame(out_p_l).to_csv(os.path.join(con.output_base_dir, save_name), header=0, index=0,
float_format='%.3f', sep=',')
print('Batch circle segmentation processing done!')
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--mode', type=str, default='batch_circle_segmentation',
help='mode optional parameters: infer_vis, infer_results, batch_circle_segmentation,'
'infer_vis: show one sample inference result; infer_results: evaluate testset results; '
'batch_circle_segmentation: output result files of the testset ')
opt = parser.parse_args()
if opt.mode == 'infer_vis':
infer_vis()
elif opt.mode == 'infer_results':
infer_results()
elif opt.mode == 'batch_circle_segmentation':
batch_circle_segmentation()