This repository has been archived by the owner on Feb 3, 2024. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 42
/
dataset.py
202 lines (165 loc) · 7.26 KB
/
dataset.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
#!/usr/bin/env python
# -*- coding: utf-8 -*-
"""
@Author: An Tao
@Contact: [email protected]
@File: dataset.py
@Time: 2020/1/2 10:26 AM
"""
import os
import torch
import json
import h5py
from glob import glob
import numpy as np
import torch.utils.data as data
shapenetpart_cat2id = {'airplane': 0, 'bag': 1, 'cap': 2, 'car': 3, 'chair': 4,
'earphone': 5, 'guitar': 6, 'knife': 7, 'lamp': 8, 'laptop': 9,
'motor': 10, 'mug': 11, 'pistol': 12, 'rocket': 13, 'skateboard': 14, 'table': 15}
shapenetpart_seg_num = [4, 2, 2, 4, 4, 3, 3, 2, 4, 2, 6, 2, 3, 3, 3, 3]
shapenetpart_seg_start_index = [0, 4, 6, 8, 12, 16, 19, 22, 24, 28, 30, 36, 38, 41, 44, 47]
def translate_pointcloud(pointcloud):
xyz1 = np.random.uniform(low=2./3., high=3./2., size=[3])
xyz2 = np.random.uniform(low=-0.2, high=0.2, size=[3])
translated_pointcloud = np.add(np.multiply(pointcloud, xyz1), xyz2).astype('float32')
return translated_pointcloud
def jitter_pointcloud(pointcloud, sigma=0.01, clip=0.02):
N, C = pointcloud.shape
pointcloud += np.clip(sigma * np.random.randn(N, C), -1*clip, clip)
return pointcloud
def rotate_pointcloud(pointcloud):
theta = np.pi*2 * np.random.rand()
rotation_matrix = np.array([[np.cos(theta), -np.sin(theta)],[np.sin(theta), np.cos(theta)]])
pointcloud[:,[0,2]] = pointcloud[:,[0,2]].dot(rotation_matrix) # random rotation (x,z)
return pointcloud
class Dataset(data.Dataset):
def __init__(self, root, dataset_name='modelnet40', class_choice=None,
num_points=2048, split='train', load_name=True, load_file=True,
segmentation=False, random_rotate=False, random_jitter=False,
random_translate=False):
assert dataset_name.lower() in ['shapenetcorev2', 'shapenetpart',
'modelnet10', 'modelnet40', 'shapenetpartpart']
assert num_points <= 2048
if dataset_name in ['shapenetcorev2', 'shapenetpart', 'shapenetpartpart']:
assert split.lower() in ['train', 'test', 'val', 'trainval', 'all']
else:
assert split.lower() in ['train', 'test', 'all']
if dataset_name not in ['shapenetpart'] and segmentation == True:
raise AssertionError
self.root = os.path.join(root, dataset_name + '_hdf5_2048')
self.dataset_name = dataset_name
self.class_choice = class_choice
self.num_points = num_points
self.split = split
self.load_name = load_name
self.load_file = load_file
self.segmentation = segmentation
self.random_rotate = random_rotate
self.random_jitter = random_jitter
self.random_translate = random_translate
self.path_h5py_all = []
self.path_name_all = []
self.path_file_all = []
if self.split in ['train', 'trainval', 'all']:
self.get_path('train')
if self.dataset_name in ['shapenetcorev2', 'shapenetpart', 'shapenetpartpart']:
if self.split in ['val', 'trainval', 'all']:
self.get_path('val')
if self.split in ['test', 'all']:
self.get_path('test')
data, label, seg = self.load_h5py(self.path_h5py_all)
if self.load_name or self.class_choice != None:
self.name = np.array(self.load_json(self.path_name_all)) # load label name
if self.load_file:
self.file = np.array(self.load_json(self.path_file_all)) # load file name
self.data = np.concatenate(data, axis=0)
self.label = np.concatenate(label, axis=0)
if self.segmentation:
self.seg = np.concatenate(seg, axis=0)
if self.class_choice != None:
indices = (self.name == class_choice)
self.data = self.data[indices]
self.label = self.label[indices]
self.name = self.name[indices]
if self.segmentation:
self.seg = self.seg[indices]
id_choice = shapenetpart_cat2id[class_choice]
self.seg_num_all = shapenetpart_seg_num[id_choice]
self.seg_start_index = shapenetpart_seg_start_index[id_choice]
if self.load_file:
self.file = self.file[indices]
elif self.segmentation:
self.seg_num_all = 50
self.seg_start_index = 0
def get_path(self, type):
path_h5py = os.path.join(self.root, '%s*.h5'%type)
paths = glob(path_h5py)
paths_sort = [os.path.join(self.root, type + str(i) + '.h5') for i in range(len(paths))]
self.path_h5py_all += paths_sort
if self.load_name:
paths_json = [os.path.join(self.root, type + str(i) + '_id2name.json') for i in range(len(paths))]
self.path_name_all += paths_json
if self.load_file:
paths_json = [os.path.join(self.root, type + str(i) + '_id2file.json') for i in range(len(paths))]
self.path_file_all += paths_json
return
def load_h5py(self, path):
all_data = []
all_label = []
all_seg = []
for h5_name in path:
f = h5py.File(h5_name, 'r+')
data = f['data'][:].astype('float32')
label = f['label'][:].astype('int64')
if self.segmentation:
seg = f['seg'][:].astype('int64')
f.close()
all_data.append(data)
all_label.append(label)
if self.segmentation:
all_seg.append(seg)
return all_data, all_label, all_seg
def load_json(self, path):
all_data = []
for json_name in path:
j = open(json_name, 'r+')
data = json.load(j)
all_data += data
return all_data
def __getitem__(self, item):
point_set = self.data[item][:self.num_points]
label = self.label[item]
if self.load_name:
name = self.name[item] # get label name
if self.load_file:
file = self.file[item] # get file name
if self.random_rotate:
point_set = rotate_pointcloud(point_set)
if self.random_jitter:
point_set = jitter_pointcloud(point_set)
if self.random_translate:
point_set = translate_pointcloud(point_set)
# convert numpy array to pytorch Tensor
point_set = torch.from_numpy(point_set)
label = torch.from_numpy(np.array([label]).astype(np.int64))
label = label.squeeze(0)
if self.segmentation:
seg = self.seg[item]
seg = torch.from_numpy(seg)
return point_set, label, seg, name, file
else:
return point_set, label, name, file
def __len__(self):
return self.data.shape[0]
if __name__ == '__main__':
root = os.getcwd()
# choose dataset name from 'shapenetcorev2', 'shapenetpart', 'modelnet40' and 'modelnet10'
dataset_name = 'shapenetcorev2'
# choose split type from 'train', 'test', 'all', 'trainval' and 'val'
# only shapenetcorev2 and shapenetpart dataset support 'trainval' and 'val'
split = 'train'
d = Dataset(root=root, dataset_name=dataset_name, num_points=2048, split=split)
print("datasize:", d.__len__())
item = 0
ps, lb, n, f = d[item]
print(ps.size(), ps.type(), lb.size(), lb.type(), n, f)