/
shapenet_edge.py
150 lines (125 loc) · 6.24 KB
/
shapenet_edge.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
# from __future__ import print_function
import torch.utils.data as data
import torch
import numpy as np
import h5py
import transforms3d
import random
import math
class PartDataset(data.Dataset):
def __init__(self, args, path,training=True):
self.npoints = args.n_in_points
self.args=args
# with h5py.File(path, 'r') as f:
f=h5py.File(path, 'r')
self.complete_pcds = f['complete_pcds'][()] # torch.from_numpy(f['complete_pcds'])#[()])
if args.n_gt_points==2048:
self.gt_pcds = self.complete_pcds # torch.from_numpy(f['complete_pcds'])#[()])
elif args.n_gt_points==16384:
self.gt_pcds = f['complete_pcds_16384'][()] # torch.from_numpy(f['complete_pcds'])#[()])
self.labels = f['labels'][()] #torch.from_numpy(f['labels'])#[()])
self.edge_labels = f['edge_labels'][()] #torch.from_numpy(f['edge_labels'])#[()])
self.edge_pcds = f['complete_edge_pcds'][()]
f.close()
if args.train_seen:
data_ids = np.isin(self.labels, np.asarray([1,2,3,4,5,6,8,9,10,11,14,15]))
self.complete_pcds = self.complete_pcds[data_ids]
self.gt_pcds = self.gt_pcds[data_ids]
self.labels = self.labels[data_ids]
self.edge_labels =self.edge_labels[data_ids]
self.edge_pcds=self.edge_pcds[data_ids]
if args.test_unseen:
data_ids = np.isin(self.labels, np.asarray([0,7,12,13]))
self.complete_pcds = self.complete_pcds[data_ids]
self.gt_pcds = self.gt_pcds[data_ids]
self.labels = self.labels[data_ids]
self.edge_labels =self.edge_labels[data_ids]
self.edge_pcds = self.edge_pcds[data_ids]
self.model_ids = torch.tensor(range(self.complete_pcds.shape[0]))
self.training=training
def __getitem__(self, index):
complete = self.complete_pcds[self.model_ids[index]]
complete=torch.from_numpy(complete)#[()])
partial,sel_ids = self.del_ratio_pts(self.args, complete, delete=self.args.remove_point_num, training=self.training)
cls = self.labels[self.model_ids[index]]
edge_labels = self.edge_labels[self.model_ids[index]]
edge_labels = torch.from_numpy(edge_labels)
edge = self.edge_pcds[self.model_ids[index]]
edge = torch.from_numpy(edge)
if self.args.n_gt_points == 16384:
complete=self.gt_pcds[self.model_ids[index]]
complete = torch.from_numpy(complete)
return partial, complete, edge_labels, cls,edge
def del_ratio_pts(self,args, batch_data, delete=512, training=True):
if training:
seed = batch_data[np.random.choice(batch_data.shape[0], 1)[0], :].unsqueeze(0)
else:
seed = batch_data[0, :].unsqueeze(0)
seed = torch.repeat_interleave(seed, batch_data.shape[0], dim=0)
diff = batch_data - seed
dist_sq = torch.sum(diff * diff, 1)
dist_sq_id = torch.argsort(dist_sq) # ascending order
sel_id = dist_sq_id[delete:]
sel_id = self.pad_cloudN(sel_id, args.n_in_points) # tmp_pt[choice, :]
return batch_data[sel_id], sel_id
def pad_cloudN(self,P, Nin):
""" Pad or subsample 3D Point cloud to Nin number of points """
N = P.shape[0]
ii = np.random.choice(N, Nin - N)
choice = np.concatenate([range(N), ii])
choice = torch.from_numpy(choice) # .to(P.device)
P = P[choice]
return P
def __len__(self):
return len(self.model_ids)
class PartDatasetPCN(data.Dataset):
def __init__(self, args, path,training=True):
self.npoints = args.n_in_points
self.args=args
f=h5py.File(path, 'r')
self.incomplete_pcds = f['incomplete_pcds']#[()]
self.complete_pcds = f['complete_pcds']#[()] # torch.from_numpy(f['complete_pcds'])##[()])
self.edge_pcds = f['complete_edge_pcds']#[()]
self.gt_pcds = self.complete_pcds
self.labels = f['labels']
self.edge_labels = f['edge_labels']#[()]
self.model_ids = torch.tensor(range(self.incomplete_pcds.shape[0]))
if args.cal_edge:
self.model_ids = self.model_ids[self.model_ids != 73479]
self.model_ids = self.model_ids[self.model_ids != 73477]
self.training=training
def augment_cloud(self,Ps, args):
"""" Augmentation on XYZ and jittering of everything """
M = transforms3d.zooms.zfdir2mat(1)
if args.pc_augm_scale > 1:
s = random.uniform(1 / args.pc_augm_scale, 1)
M = np.dot(transforms3d.zooms.zfdir2mat(s), M)
if args.pc_augm_rot > 0:
angle = random.uniform(0, math.pi / 180 * args.pc_augm_rot)
M = np.dot(transforms3d.axangles.axangle2mat([0, 1, 0], angle), M) # y=upright assumption
if args.pc_augm_mirror_prob > 0: # mirroring x&z, not y
if random.random() < args.pc_augm_mirror_prob / 2:
M = np.dot(transforms3d.zooms.zfdir2mat(-1, [np.random.random() - 0.5, 0, np.random.random() - 0.5]), M)
result = []
for P in Ps:
tmp = np.dot(P[:, :3], M.T)
if args.pc_augm_jitter:
sigma, clip = 0.01, 0.05 # https://github.com/charlesq34/pointnet/blob/master/provider.py#L74
tmp = tmp + np.clip(sigma * np.random.randn(*P.shape), -1 * clip, clip).astype(np.float32)
result.append(tmp)
return result
def __getitem__(self, index):
complete=self.gt_pcds[self.model_ids[index]]
edge = self.edge_pcds[self.model_ids[index]]
partial = self.incomplete_pcds[self.model_ids[index]]
if self.args.augment and self.training:
complete, partial,edge = self.augment_cloud([complete, partial,edge], self.args)
complete=torch.from_numpy(complete).float()#.clamp(-0.5, 0.5)
edge = torch.from_numpy(edge).float()#.clamp(-0.5, 0.5)
partial = torch.from_numpy(partial).float().clamp(-self.args.normalize_ratio, self.args.normalize_ratio)
cls = self.labels[self.model_ids[index]]
edge_labels = self.edge_labels[self.model_ids[index]]
edge_labels = torch.from_numpy(edge_labels)
return partial, complete, edge_labels, cls, edge
def __len__(self):
return len(self.model_ids)