-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathdataset.py
executable file
·225 lines (203 loc) · 11.6 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
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
import torch.utils.data as data
import json
import random
from PIL import Image
import numpy as np
import torch
import os
def generate_class_info(dataset_name):
class_name_map_class_id = {}
if dataset_name == 'mvtec_pc_3d_rgb':
obj_list = ["bagel", "cable_gland", "carrot", "cookie", "dowel", "foam", "peach", "potato", "rope", "tire",]
elif dataset_name == 'eye_pc_3d_rgb':
obj_list = [
'CandyCane',
'ChocolateCookie',
'ChocolatePraline',
'Confetto',
'GummyBear',
'HazelnutTruffle',
'LicoriceSandwich',
'Lollipop',
'Marshmallow',
'PeppermintCandy']
elif dataset_name == 'real_pc_3d_rgb':
obj_list = ['airplane','car','candybar','chicken',
'diamond','duck','fish','gemstone',
'seahorse','shell','starfish','toffees']
for k, index in zip(obj_list, range(len(obj_list))):
class_name_map_class_id[k] = index
return obj_list, class_name_map_class_id
def generate_is_seen_for_point_in_each_view(d2_3d_cor_list, non_zero_index_arr):
# nv, l, 3
d2_3d_cor = torch.stack(d2_3d_cor_list)
# nv, l
is_seen = d2_3d_cor[..., 2]
nonzero_index = np.nonzero(np.asarray(non_zero_index_arr).reshape(-1,))[0]
max_nonzero_index, min_nonzero_index = nonzero_index[-1], nonzero_index[0]
mask = np.asarray(is_seen.bool() | (~non_zero_index_arr.permute(1, 0).bool()))
# Few points have no projections in any views, we fill the d2_3d_cor of its neighbors as its projection to avoid the nan value when back-projecting 2d presentations to 3d
is_seen = np.nonzero(np.all(mask == 0, axis = 0))[0]
if len(is_seen):
for i in is_seen:
success_find = 0
up_select_idx = 0
down_select_idx = 0
for j in range(i+1, max_nonzero_index + 1):
up_select_idx = j
if np.any(d2_3d_cor[..., j, 2].numpy()):
success_find = 1
dis = np.abs(up_select_idx - i)
select_index = up_select_idx
break
if not success_find:
for j in range(min_nonzero_index, i):
up_select_idx = j
if np.any(d2_3d_cor[..., j, 2].numpy()):
success_find = 1
dis = np.abs(max_nonzero_index - i + up_select_idx - min_nonzero_index)
select_index = up_select_idx
break
if not success_find:
raise NotImplementedError("bug")
for j in reversed(range(min_nonzero_index, i)):
down_select_idx = j
if dis <= np.abs(down_select_idx - i):
select_index = up_select_idx
break;
else:
if np.any(d2_3d_cor[..., j, 2].numpy()):
success_find = 1
select_index = down_select_idx
break
if not success_find:
for j in reversed(range(i+1, max_nonzero_index + 1)):
down_select_idx = j
if dis <= np.abs(i - min_nonzero_index + max_nonzero_index - down_select_idx):
select_index = up_select_idx
break;
else:
if np.any(d2_3d_cor[..., j, 2].numpy()):
success_find = 1
select_index = down_select_idx
break
if not success_find:
raise NotImplementedError("bug")
d2_3d_cor[..., i, 2]=d2_3d_cor[..., select_index, 2]
is_seen = d2_3d_cor[..., 2]
mask = np.asarray(is_seen.bool() | (~non_zero_index_arr.permute(1, 0).bool()))
is_seen = np.nonzero(np.all(mask == 0, axis = 0))[0]
return is_seen, d2_3d_cor
import open3d
class Dataset(data.Dataset):
def __init__(self, root, transform, target_transform, target_transform_pc, dataset_name, train_dataset_name = None, point_size = 336, is_all = False, mode='test'):
self.root = root
self.transform = transform
self.target_transform = target_transform
self.target_transform_pc = target_transform_pc
self.point_size = point_size
self.dataset_name = dataset_name
self.data_all = []
if is_all:
meta_info = json.load(open(f'{self.root}/all_meta.json', 'r'))
else:
if mode == 'test':
meta_info = json.load(open(f'{self.root}/{train_dataset_name}_meta.json', 'r'))
else:
meta_info = json.load(open(f'{self.root}/{obj_name}_meta.json', 'r'))
# meta_info = json.load(open(f'{self.root}/meta.json', 'r'))
# meta_info = json.load(open(f'/remote-home/iot_zhouqihang/data/mvtec_3d_mv/mvtec_3d_9_views/meta_test.json', 'r'))
name = self.root.split('/')[-1]
meta_info = meta_info[mode]
self.cls_names = list(meta_info.keys())
for cls_name in self.cls_names:
self.data_all.extend(meta_info[cls_name])
self.length = len(self.data_all)
self.obj_list, self.class_name_map_class_id = generate_class_info(dataset_name)
def __len__(self):
return self.length
def __getitem__(self, index):
data = self.data_all[index]
if self.dataset_name == 'mvtec_pc_3d_rgb' or self.dataset_name == 'eye_pc_3d_rgb':
img_path, mask_path, cls_name, specie_name, anomaly = data['d2_img_path'], data['d2_mask_path'], data['cls_name'], \
data['specie_name'], data['anomaly']
d2_render_img_path, d2_render_gt_path, d2_corrdinate_path = data['d2_render_img_path'], data['d2_render_gt_path'], data['d2_corrdinate']
elif self.dataset_name == 'real_pc_3d_rgb':
mask_path, cls_name, specie_name, anomaly = data['d2_mask_path'], data['cls_name'], \
data['specie_name'], data['anomaly']
d2_render_img_path, d2_render_gt_path, d2_corrdinate_path = data['d2_render_img_path'], data['d2_render_gt_path'], data['d2_corrdinate']
# load 2d rendering images
d2_render_img_path_list = []
for filename in sorted(os.listdir(d2_render_img_path)):
img = Image.open(os.path.join(d2_render_img_path, filename)).convert("RGB")
img = self.transform(img) if self.transform is not None else img
d2_render_img_path_list.append(img)
d2_render_img = torch.stack(d2_render_img_path_list)
# load 2d rendering groundtruth
d2_render_gt_path_list = []
rendering_anomaly_list = []
for filename in sorted(os.listdir(d2_render_gt_path)):
img_mask = Image.open((os.path.join(d2_render_gt_path, filename))).convert('L')
img_mask = self.target_transform(img_mask)
img_mask[img_mask>0.5] = 1.0
img_mask[img_mask<=0.5] = 0.0
rendering_anomaly = 0.0 if torch.all(img_mask == 0) else 1.0
d2_render_gt_path_list.append(img_mask)
rendering_anomaly_list.append(rendering_anomaly)
d2_render_gt = torch.stack(d2_render_gt_path_list)
rendering_anomaly = torch.tensor(rendering_anomaly_list)
# load the correspondence between points and pixels in each view
d2_3d_cor_list = []
non_zero_index_list = []
# for organized point ckoud
if self.dataset_name == 'mvtec_pc_3d_rgb' or self.dataset_name == 'eye_pc_3d_rgb':
for idx, filename in enumerate(sorted(os.listdir(d2_corrdinate_path))):
if idx == 0:
template_non_zero_index = torch.zeros(self.point_size * self.point_size, dtype = torch.long)
non_zero_index = np.load((os.path.join(d2_corrdinate_path, filename)))
non_zero_index = torch.from_numpy(non_zero_index)
template_non_zero_index[non_zero_index] = 1
non_zero_index_arr = template_non_zero_index.reshape(-1, 1)
else:
template_d2_corrdinate = torch.zeros(self.point_size * self.point_size, 3, dtype = torch.long)
d2_corrdinate = np.load((os.path.join(d2_corrdinate_path, filename)))
d2_corrdinate = torch.from_numpy(d2_corrdinate).long()
template_d2_corrdinate[non_zero_index] = d2_corrdinate
d2_3d_cor_list.append(template_d2_corrdinate)
# for unorganized point cloud
elif self.dataset_name == 'real_pc_3d_rgb':
for idx, filename in enumerate(sorted(os.listdir(d2_corrdinate_path))):
if idx == 0:
template_non_zero_index = torch.ones(self.point_size * self.point_size, dtype = torch.long)
non_zero_index_arr = template_non_zero_index.reshape(-1, 1)
d2_corrdinate = np.load((os.path.join(d2_corrdinate_path, filename)))
d2_corrdinate = torch.from_numpy(d2_corrdinate).long()
template_d2_corrdinate = d2_corrdinate
d2_3d_cor_list.append(template_d2_corrdinate)
# remove the hidden points in each view
is_seen, d2_3d_cor = generate_is_seen_for_point_in_each_view(d2_3d_cor_list, non_zero_index_arr)
if self.dataset_name == 'mvtec_pc_3d_rgb' or self.dataset_name == 'eye_pc_3d_rgb':
img = Image.open(os.path.join(self.root, img_path))
if anomaly == 0:
img_mask = Image.fromarray(np.zeros((img.size[0], img.size[1])), mode='L')
else:
if os.path.isdir(os.path.join(self.root, mask_path)):
img_mask = Image.fromarray(np.zeros((img.size[0], img.size[1])), mode='L')
else:
img_mask = np.array(Image.open(os.path.join(self.root, mask_path)).convert('L')) > 0
img_mask = Image.fromarray(img_mask.astype(np.uint8) * 255, mode='L')
# transforms
img = self.transform(img) if self.transform is not None else img
img_mask = self.target_transform(
img_mask) if self.target_transform is not None and img_mask is not None else img_mask
img_mask = [] if img_mask is None else img_mask
return {'img': img, 'img_mask': img_mask, 'cls_name': cls_name, 'anomaly': anomaly, 'd2_render_img': d2_render_img, 'd2_render_anomaly': rendering_anomaly, 'd2_render_gt': d2_render_gt, 'd2_3d_cor': d2_3d_cor,
'img_path': os.path.join(self.root, img_path), "cls_id":self.class_name_map_class_id[cls_name], "d2_render_img_path": d2_render_img_path, "non_zero_index": non_zero_index_arr, "index":torch.LongTensor([index])}
elif self.dataset_name == 'real_pc_3d_rgb':
pcd = open3d.io.read_point_cloud(os.path.join(self.root, mask_path))
img_mask = np.array(pcd.colors)
img_mask[img_mask>0.5] = 1.0
img_mask[img_mask<0.5] = 0.0
img_mask = np.all(img_mask, axis = 1).astype(int)
return {'img': '', 'img_mask': img_mask, 'cls_name': cls_name, 'anomaly': anomaly, 'd2_render_img': d2_render_img, 'd2_render_anomaly': rendering_anomaly, 'd2_render_gt': d2_render_gt, 'd2_3d_cor': d2_3d_cor,
'img_path': '', "cls_id":self.class_name_map_class_id[cls_name], "d2_render_img_path": d2_render_img_path, "non_zero_index": non_zero_index_arr}