/
test_vis_attn_map_3d.py
165 lines (129 loc) · 6.02 KB
/
test_vis_attn_map_3d.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
import argparse
import os
import sys
import open3d as o3d
import numpy as np
import sklearn.neighbors
import torch
import torch.nn.parallel
from tqdm import tqdm
from utils.pcp_name_filter import get_pt_clouds_path
from utils.training_utils import set_randomness, load_ckpt_to_net
from metrics import comp_rms_angle_batch
from model import NINormalNet
def norm_pts_to_unit_sphere(pts):
"""
:param pts: N, 3
:return:
pts: N, 3
radius: scalar
"""
pts_range = np.max(pts, axis=0) - np.min(pts, axis=0)
max_range = np.max(pts_range) # this is the max diameter
radius = max_range / 2.0
pts /= radius
return pts, radius
def parse_args():
parser = argparse.ArgumentParser('NINormalNetTestVis3DPatch')
parser.add_argument('--gpu_id', default=0, type=int)
parser.add_argument('--multi_gpu', default=True, type=bool)
parser.add_argument('--datafolder', type=str, default='./dataset_dir/PCPNet_official_dataset',
help='folder contains h5 dataset')
parser.add_argument('--num_neighbours', type=int, default=50)
parser.add_argument('--datatype', type=str, default='test')
parser.add_argument('--truerand', type=bool, default=False, help="whether we want true randomness")
parser.add_argument('--randseed', default=20, help="set random seed for np, python, and torch")
parser.add_argument('--fastdebug', default=False, action='store_true', help="debug with very small portion data")
parser.add_argument('--ckpt_path', type=str, help="checkpoint_folder",
default='./paper_ckpts/nb50')
return parser.parse_args()
def test_one_epoch(model):
model.eval()
K = args.num_neighbours
noise_type = 'none'
noise_intensity = '0.0'
obj_names = None
if args.datatype is 'train':
obj_names = np.genfromtxt(os.path.join(args.datafolder, 'trainingset_no_noise.txt'), dtype='str')
elif args.datatype is 'test':
obj_names = np.genfromtxt(os.path.join(args.datafolder, 'testset_no_noise.txt'), dtype='str')
elif args.datatype is 'eval':
obj_names = np.genfromtxt(os.path.join(args.datafolder, 'validationset_no_noise.txt'), dtype='str')
obj_paths = get_pt_clouds_path(args.datafolder, obj_names, noise_type=noise_type, noise_intensity=noise_intensity)
obj_pts_files = [p + '.xyz' for p in obj_paths]
obj_normals_files = [p + '.normals' for p in obj_paths]
for i in range(len(obj_pts_files)):
obj_name = obj_names[i]
# Uncomment this to visualise a specific object.
# if obj_name != 'netsuke100k':
# continue
print(noise_type, obj_name, obj_pts_files[i])
# shift the centre of the point cloud to origin, and normalise the entire point cloud to unit sphere
pts = np.genfromtxt(obj_pts_files[i])
pts = pts - np.mean(pts, axis=0)
pts, _ = norm_pts_to_unit_sphere(pts)
normals = np.genfromtxt(obj_normals_files[i])
N = pts.shape[0]
tree = sklearn.neighbors.KDTree(pts[:N], leaf_size=50)
# this controls how many patches we would like to vis for each object.
counter = 0
max_count = 7
for r in tqdm(range(N)):
'''Estimate normal and attn weights for a patch'''
# get neighbours and shift to the origin
_, idx = tree.query(pts[r].reshape(1, 3), k=K) # the first one is itself
knn = pts[idx.squeeze()]
normal_gt = normals[idx.squeeze()][0] # (3, )
centroid = np.mean(knn, axis=0)
knn_centred = knn - centroid
knn_model_input = torch.from_numpy(knn_centred) # (K, 3)
knn_model_input = knn_model_input.view(1, K, 1, 3) # (B, K, N, 3)
knn_model_input = knn_model_input.cuda().float()
# pred_normals: (B, N, 3), weights: (N, K)
pred_normals, weights = model(knn_model_input) # (1, 1, 3), (1, 50)
normal_gt = torch.from_numpy(normal_gt).float().view(1, 1, 3).cuda()
rms_angle = comp_rms_angle_batch(pred_normals, normal_gt) # (1, 1)
tqdm.write('Angle err {0:.4f}'.format(rms_angle))
'''This is vis the patch in the entire point cloud'''
pcd_entire = o3d.geometry.PointCloud()
color = np.zeros_like(pts)
color[:, 1] = 0.3 # all points are green
color[idx.squeeze(), 1] = 0
color[idx.squeeze(), 0] = 1 # selected points are red
color[r] = 0
color[r, 2] = 1 # the point is blue
pcd_entire.points = o3d.utility.Vector3dVector(pts)
pcd_entire.colors = o3d.utility.Vector3dVector(color)
pcd_entire.normals = o3d.utility.Vector3dVector(normals)
o3d.visualization.draw_geometries([pcd_entire])
'''This is just vis the patch'''
pcd_knn = o3d.geometry.PointCloud()
color = np.zeros_like(knn_centred)
# we need to amplify all attention weights, otherwise the colour-coding looks dull.
max_weight = weights.max().item()
enlarge_ratio = 1.0 / max_weight
color[:, 0] = weights.squeeze().cpu().numpy() * enlarge_ratio * 7 # selected points are red
color[0] = 0
color[0, 2] = 1 # the point is blue
pcd_knn.points = o3d.utility.Vector3dVector(knn_centred)
pcd_knn.colors = o3d.utility.Vector3dVector(color)
o3d.visualization.draw_geometries([pcd_knn])
if counter >= max_count:
break
counter += 1
def main(args):
'''Model Loading'''
ckpt_file = os.path.join(args.ckpt_path, 'ni_normal_net.pth')
model = NINormalNet()
if args.multi_gpu:
model = torch.nn.DataParallel(model).to(device='cuda:' + str(args.gpu_id))
else:
model = model.to(device='cuda:'+str(args.gpu_id))
load_ckpt_to_net(ckpt_file, model)
'''Testing'''
test_one_epoch(model)
if __name__ == '__main__':
args = parse_args()
set_randomness(args)
with torch.no_grad():
main(args)