-
Notifications
You must be signed in to change notification settings - Fork 11
/
scannet_dataset.py
executable file
·340 lines (294 loc) · 15.9 KB
/
scannet_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
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
import SharedArray as SA
import os
import numpy as np
import glob
import torch
import pickle
import json
import cv2
import copy
from torch.utils import data
from ..indoor_dataset import IndoorDataset
from ...utils.common_utils import sa_create, sa_delete
class ScanNetDataset(IndoorDataset):
def __init__(self, dataset_cfg, class_names, training, root_path, logger=None):
super(ScanNetDataset, self).__init__(
dataset_cfg, class_names, training, root_path, logger=logger
)
self.data_suffix = dataset_cfg.DATA_SPLIT.data_suffix
self.data_list = sorted(
glob.glob(str(self.root_path / dataset_cfg.DATA_SPLIT[self.mode]) + '/*' + self.data_suffix))
self.split_file = dataset_cfg.DATA_SPLIT[self.mode]
self.put_data_to_shm()
if self.training and hasattr(self, 'caption_cfg') and self.caption_cfg.get('CAPTION_CORR_PATH_IN_ONE_FILE', True):
self.scene_image_corr_infos, self.scene_image_corr_entity_infos = self.include_caption_infos()
self.load_image = self.dataset_cfg.get('LOAD_IMAGE', None)
self.depth_image_scale = self.dataset_cfg.get('DEPTH_IMAGE_SCALE', None)
self.image_path = self.dataset_cfg.get('IMAGE_PATH', None)
if self.load_image:
self.pc_means = json.load(open(str(self.root_path / 'pc_mean.json')))
self.logger.info(
"Totally {} samples in {} set.".format(
len(self.data_list) * (self.repeat if self.training else 1), self.mode))
def __len__(self):
length = len(self.data_list) * (self.repeat if self.training else 1)
if self._merge_all_iters_to_one_epoch:
return length * self.total_epochs
else:
return length
def put_data_to_shm(self):
n_classes = self.n_classes
for item in self.data_list:
if self.cache and not os.path.exists("/dev/shm/scannet_{}".format(item.split('/')[-1][:-4] + '_xyz_{}'.format(n_classes))):
if self.split_file.find('test') < 0:
xyz, rgb, label, inst_label, *others = torch.load(item)
sa_create("shm://scannet_{}".format(item.split('/')[-1][:-4] + '_label_{}'.format(n_classes)), label)
sa_create("shm://scannet_{}".format(item.split('/')[-1][:-4] + '_inst_label_{}'.format(n_classes)), inst_label)
else:
xyz, rgb = torch.load(item)
sa_create("shm://scannet_{}".format(item.split('/')[-1][:-4] + '_xyz_{}'.format(n_classes)), xyz)
sa_create("shm://scannet_{}".format(item.split('/')[-1][:-4] + '_rgb_{}'.format(n_classes)), rgb)
def load_data(self, index):
n_classes = self.n_classes
fn = self.data_list[index]
if self.cache:
xyz = SA.attach("shm://scannet_{}".format(fn.split('/')[-1][:-4] + '_xyz_{}'.format(n_classes))).copy()
rgb = SA.attach("shm://scannet_{}".format(fn.split('/')[-1][:-4] + '_rgb_{}'.format(n_classes))).copy()
if self.split_file.find('test') < 0:
label = SA.attach("shm://scannet_{}".format(fn.split('/')[-1][:-4] + '_label_{}'.format(n_classes))).copy()
inst_label = SA.attach("shm://scannet_{}".format(fn.split('/')[-1][:-4] + '_inst_label_{}'.format(n_classes))).copy()
else:
label = np.full(xyz.shape[0], self.ignore_label).astype(np.int64)
inst_label = np.full(xyz.shape[0], self.ignore_label).astype(np.int64)
else:
if self.split_file.find('test') < 0:
xyz, rgb, label, inst_label, *others = torch.load(fn)
else:
xyz, rgb = torch.load(fn)
label = np.full(xyz.shape[0], self.ignore_label)
inst_label = np.full(xyz.shape[0], self.ignore_label)
# base / novel label
if hasattr(self, 'base_class_mapper'):
binary_label = self.binary_class_mapper[label.astype(np.int64)].astype(np.float32)
else:
binary_label = np.ones_like(label)
if self.class_mode == 'base':
label = self.base_class_mapper[label.astype(np.int64)]
elif self.class_mode == 'all' and hasattr(self, 'ignore_class_idx'):
label = self.valid_class_mapper[label.astype(np.int64)]
inst_label[label == self.ignore_label] = self.ignore_label
return xyz, rgb, label, inst_label, binary_label
def __getitem__(self, item):
index = item % len(self.data_list)
xyz, rgb, label, inst_label, binary_label = self.load_data(index)
pc_count = xyz.shape[0]
origin_idx = np.arange(xyz.shape[0]).astype(np.int64)
# === caption ===
scene_name = self.data_list[index].split('/')[-1].split('.')[0]
# get captioning data
if self.training and hasattr(self, 'caption_cfg'):
if self.caption_cfg.get('CAPTION_CORR_PATH_IN_ONE_FILE', True):
image_corr_dict, image_name_dict = self.get_caption_image_corr_and_name_from_memory(scene_name, index)
else:
image_corr_dict, image_name_dict = self.get_caption_image_corr_and_name_from_file(scene_name)
caption_data = self.select_caption_and_idx_all(scene_name, image_name_dict, image_corr_dict)
else:
caption_data = None
if not self.rgb_norm:
rgb = (rgb + 1) * 127.5
data_dict = {
'points_xyz': xyz, 'rgb': rgb, 'labels': label, 'inst_label': inst_label,
'binary_labels': binary_label, 'origin_idx': origin_idx, 'pc_count': pc_count,
'caption_data': caption_data, 'ids': index, 'scene_name': scene_name
}
# === load images ===
if self.load_image:
info = {'scene_name': scene_name, 'depth_image_size': self.depth_image_scale}
data_dict = self.get_image(info, data_dict)
if self.training:
# perform augmentations
data_dict = self.augmentor.forward(data_dict)
if not data_dict['valid']:
return ScanNetDataset.__getitem__(self, np.random.randint(self.__len__()))
else:
xyz_voxel_scale = xyz * self.voxel_scale
xyz_voxel_scale -= xyz_voxel_scale.min(0)
data_dict['points_xyz_voxel_scale'] = xyz_voxel_scale
data_dict['points'] = xyz
# prepare features for voxelization
if self.dataset_cfg.DATA_PROCESSOR.rgb_as_feat:
data_dict['feats'] = data_dict['rgb']
if self.dataset_cfg.DATA_PROCESSOR.xyz_as_feat:
if 'feats' in data_dict:
data_dict['feats'] = np.concatenate((data_dict['feats'], data_dict['points_xyz']), axis=1)
else:
data_dict['feats'] = data_dict['points_xyz']
data_dict = self.data_processor.forward(data_dict)
# visualization debug code
# import tools.visual_utils.open3d_vis_utils as vis
# vis_dict = {
# 'points': data_dict['points'],
# 'point_colors': data_dict['rgb'],
# 'point_size': 2.0
# }
# vis.dump_vis_dict(vis_dict)
# import ipdb;
# ipdb.set_trace(context=20)
# data_dict.pop('points_xyz')
return data_dict
def get_image(self, info, data_dict, resize_debug=False):
data_dict['point_img_1d'] = {}
data_dict['point_img'] = {}
data_dict['point_img_idx'] = {}
data_dict['image_shape'] = {}
scene_name = data_dict['scene_name']
depth_image_size = info['depth_image_size']
pose_paths = sorted(
glob.glob(str(self.root_path / self.image_path / scene_name / 'pose/*.txt')),
key=lambda a: os.path.basename(a).split('.')
)
depth_paths = sorted(
glob.glob(str(self.root_path / self.image_path / scene_name / 'depth/*.png')),
key=lambda a: os.path.basename(a).split('.')
)
color_paths = sorted(
glob.glob(str(self.root_path / self.image_path / scene_name / 'color/*.jpg')),
key=lambda a: os.path.basename(a).split('.')
)
assert len(pose_paths) == len(depth_paths) and len(pose_paths) == len(color_paths)
try:
depth_intrinsic = np.loadtxt(
os.path.join(self.root_path, self.image_path, scene_name, 'intrinsics_depth.txt')
)
except:
depth_intrinsic = np.loadtxt(
os.path.join(self.root_path, self.image_path, scene_name, 'intrinsic_depth.txt')
)
points_xyz = data_dict['points_xyz'] + np.array(self.pc_means[scene_name]).reshape(-1, 3)
if resize_debug:
target_size = (120, 160)
scale = (480 - 1) / (target_size[0] - 1)
depth_intrinsic[:1, :] = depth_intrinsic[:1, :] / scale
depth_intrinsic[1:2, :] = depth_intrinsic[1:2, :] / scale
depth_image_size = target_size
if depth_image_size[0] != 480:
scale = (480 - 1) / (depth_image_size[0] - 1)
depth_intrinsic[:1, :] = depth_intrinsic[:1, :] / scale
depth_intrinsic[1:2, :] = depth_intrinsic[1:2, :] / scale
for ind, (pose, depth, color) in enumerate(zip(pose_paths, depth_paths, color_paths)):
image_name = pose.split('/')[-1].split('.')[0]
point_idx, image_idx_1d, image_idx, color_image_shape = self.project_point_to_image(
points_xyz, pose, depth, color, depth_image_size, depth_intrinsic)
data_dict['point_img_1d'][image_name.lower()] = image_idx_1d
data_dict['point_img'][image_name.lower()] = image_idx
data_dict['point_img_idx'][image_name.lower()] = point_idx
data_dict['image_shape'][image_name.lower()] = color_image_shape
data_dict['depth_image_size'] = depth_image_size
return data_dict
@staticmethod
def project_point_to_image(points_world, pose_path, depth_path, color_path, image_size, depth_intrinsic):
fx = depth_intrinsic[0, 0]
fy = depth_intrinsic[1, 1]
cx = depth_intrinsic[0, 2]
cy = depth_intrinsic[1, 2]
bx = depth_intrinsic[0, 3]
by = depth_intrinsic[1, 3]
# == processing depth ===
depth_img = cv2.imread(depth_path, -1) # read 16bit grayscale image
depth_shift = 1000.0
depth = depth_img / depth_shift
depth_mask = (depth_img != 0)
# == processing color ===
color_image = cv2.imread(color_path)
color_image_shape = color_image.shape
color_image = cv2.resize(color_image, (image_size[1], image_size[0]))
# color_image = np.reshape(color_image[mask], [-1,3]) ##########
color_image = np.reshape(color_image, [-1, 3])
colors = np.zeros_like(color_image)
colors[:, 0] = color_image[:, 2]
colors[:, 1] = color_image[:, 1]
colors[:, 2] = color_image[:, 0]
# == processing pose ===
pose = np.loadtxt(pose_path)
# == 3D to camera coordination ===
points = np.hstack((points_world[..., :3], np.ones((points_world.shape[0], 1))))
points = np.dot(points, np.linalg.inv(np.transpose(pose)))
# == camera to image coordination ===
u = (points[..., 0] - bx) * fx / points[..., 2] + cx
v = (points[..., 1] - by) * fy / points[..., 2] + cy
d = points[..., 2]
u = (u + 0.5).astype(np.int32)
v = (v + 0.5).astype(np.int32)
# debug test
# img = np.zeros((image_size[0], image_size[1], 3))
# for jj in range(image_size[0]):
# for kk in range(image_size[1]):
# curr_d = d[(u == kk) & (v == jj) & (d >= 0)]
# if curr_d.shape[0] > 0:
# idx = curr_d.argmin()
# img[jj, kk] = points_world[..., 3:6][..., ::-1][(u == kk) & (v == jj) & (d >= 0)][idx]
# else:
# img[jj, kk] = 255.0
# cv2.imwrite('temp.png', img)
# import ipdb; ipdb.set_trace()
# filter out invalid points
point_valid_mask = (d >= 0) & (u < image_size[1]) & (v < image_size[0]) & (u >= 0) & (v >= 0)
point_valid_idx = np.where(point_valid_mask)[0]
point2image_coords = v * image_size[1] + u
valid_point2image_coords = point2image_coords[point_valid_idx]
depth = depth.reshape(-1)
depth_mask = depth_mask.reshape(-1)
# u_, v_ = np.meshgrid(np.linspace(0, image_size[1] - 1, image_size[1]), np.linspace(0, image_size[0] - 1, image_size[0]))
# image_coords = (v_ * image_size[1] + u_).reshape(-1)
image_depth = depth[valid_point2image_coords.astype(np.int64)]
depth_mask = depth_mask[valid_point2image_coords.astype(np.int64)]
point2image_depth = d[point_valid_idx]
depth_valid_mask = depth_mask & (np.abs(image_depth - point2image_depth) <= 0.2 * image_depth)
# depth_valid_idx = np.where(depth_valid_mask)[0] # corresponding image coords
point2image_coords_1d = valid_point2image_coords[depth_valid_mask] # corresponding image coords
point2image_coords_u = point2image_coords_1d % image_size[1] # (width, long)
point2image_coords_v = point2image_coords_1d // image_size[1] # (height, short)
point2image_coords_2d = np.concatenate([point2image_coords_u[:, None], point2image_coords_v[:, None]], axis=-1)
point_valid_idx = point_valid_idx[depth_valid_mask] # corresponding point idx
return point_valid_idx, point2image_coords_1d, point2image_coords_2d, color_image_shape
def __del__(self):
if not self.cache:
return
for item in self.data_list:
if os.path.exists("/dev/shm/scannet_{}".format(item.split('/')[-1][:-4] + '_xyz')):
sa_delete("shm://scannet_{}".format(item.split('/')[-1][:-4] + '_rgb'))
sa_delete("shm://scannet_{}".format(item.split('/')[-1][:-4] + '_xyz'))
if self.split_file.find('test') < 0:
sa_delete("shm://{}".format(item.split('/')[-1][:-4] + '_label'))
class ScanNetInstDataset(ScanNetDataset):
def __init__(self, dataset_cfg, class_names, training, root_path, logger=None):
ScanNetDataset.__init__(self, dataset_cfg, class_names, training, root_path, logger=logger)
self.inst_class_idx = dataset_cfg.inst_class_idx
self.inst_label_shift = dataset_cfg.inst_label_shift
if 'base_inst_class_idx' in dataset_cfg:
self.base_inst_class_idx = dataset_cfg.base_inst_class_idx
self.novel_inst_class_idx = dataset_cfg.novel_inst_class_idx
elif 'base_class_idx' in dataset_cfg:
self.base_inst_class_idx = np.array(self.base_class_idx)[dataset_cfg.inst_label_shift:] - self.inst_label_shift
self.novel_inst_class_idx = np.array(self.novel_class_idx) - self.inst_label_shift
self.sem2ins_classes = dataset_cfg.sem2ins_classes
self.NYU_ID = (3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 14, 16, 24, 28, 33, 34, 36, 39)
def __getitem__(self, item):
data_dict = super().__getitem__(item)
# get instance infos
# info = self.get_instance_info(xyz_mid, inst_label.astype(np.int32), label)
label, inst_label, binary_label = data_dict['labels'], data_dict['inst_label'], data_dict['binary_labels']
points = data_dict['points_xyz']
if self.training:
inst_label[binary_label == 0] = self.ignore_label
inst_label = self.get_valid_inst_label(inst_label, label != self.ignore_label)
if self.training and inst_label.max() < 0:
return ScanNetInstDataset.__getitem__(self, np.random.randint(self.__len__()))
info = self.get_inst_info(points, inst_label.astype(np.int32), label)
data_dict['inst_label'] = inst_label
data_dict.update(info)
return data_dict
def get_inst_info(self, xyz, instance_label, semantic_label):
ret = super().get_inst_info(xyz, instance_label, semantic_label)
ret['inst_cls'] = [x - self.inst_label_shift if x != -100 else x for x in ret['inst_cls']]
return ret