-
Notifications
You must be signed in to change notification settings - Fork 11
/
dataio.py
390 lines (297 loc) · 14.8 KB
/
dataio.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
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
import os
import numpy as np
import torch
from torch.utils.data import Dataset
import MinkowskiEngine as ME
import math
import yaml
import open3d as o3d
import random
import json
import scipy.linalg
from scipy import spatial
from scipy import ndimage
from PIL import Image
from IPython import embed
SPLIT_SEQUENCES = {
'train': ['00', '01', '02', '03', '04', '05', '06', '07', '09', '10'],
'valid': ['08'],
'test': ['11', '12', '13', '14', '15', '16', '17', '18', '19', '20', '21']
}
SPLIT_FILES = {
'train': ['.bin', '.label', '.invalid'],
'valid': ['.bin', '.label', '.invalid'],
'test': ['.bin']
}
EXT_TO_NAME = {'.bin': 'input', '.label': 'label', '.invalid': 'invalid'}
VOXEL_SCALE = [256,256,32]
config_file = os.path.join('semantic-kitti.yaml')
kitti_config = yaml.safe_load(open(config_file, 'r'))
remapdict = kitti_config['learning_map']
def unpack(compressed):
''' given a bit encoded voxel grid, make a normal voxel grid out of it. '''
uncompressed = np.zeros(compressed.shape[0] * 8, dtype=np.uint8)
uncompressed[::8] = compressed[:] >> 7 & 1
uncompressed[1::8] = compressed[:] >> 6 & 1
uncompressed[2::8] = compressed[:] >> 5 & 1
uncompressed[3::8] = compressed[:] >> 4 & 1
uncompressed[4::8] = compressed[:] >> 3 & 1
uncompressed[5::8] = compressed[:] >> 2 & 1
uncompressed[6::8] = compressed[:] >> 1 & 1
uncompressed[7::8] = compressed[:] & 1
return uncompressed
def get_eval_mask(labels, invalid_voxels):
'''
Ignore labels set to 255 and invalid voxels (the ones never hit by a laser ray, probed using ray tracing)
:param labels: input ground truth voxels
:param invalid_voxels: voxels ignored during evaluation since the lie beyond the scene that was captured by the laser
:return: boolean mask to subsample the voxels to evaluate
'''
masks = np.ones_like(labels, dtype=np.bool)
masks[labels == 255] = False
masks[invalid_voxels == 1] = False
return masks
def data_augmentation(raw, label, invalid, config):
'''flip & rotate'''
if config['DATA_IO']['augmentation']:
angle = config['DATA_IO']['augmentation_angle']
theta = random.randint(0, angle * 2) - angle
if config['DATA_IO']['augmentation_flip'] and np.random.rand(1) > 0.5:
raw = np.flip(raw, axis=1)
label = np.flip(label, axis=1)
invalid = np.flip(invalid, axis=1)
else:
theta = 0
raw = ndimage.rotate(raw, theta, reshape=False, order=0, mode='constant', cval=0)
label = ndimage.rotate(label, theta, reshape=False, order=0, mode='constant', cval=0)
invalid = ndimage.rotate(invalid, theta, reshape=False, order=0, mode='constant', cval=1)
# raw points
raw_pos = raw.nonzero()
raw_points = np.transpose(raw_pos)
# label points
label_pos = label.nonzero()
label_value = label[label_pos]
label_points = np.transpose(label_pos)
pcd = o3d.geometry.PointCloud()
pcd.points = o3d.utility.Vector3dVector(label_points[:, :3])
pcd.estimate_normals(search_param=o3d.geometry.KDTreeSearchParamKNN(knn=30))
pcd.orient_normals_to_align_with_direction(orientation_reference=np.array([0.,0.,1.]))
occupancy_normal = np.concatenate((np.asarray(pcd.points),np.asarray(pcd.normals)),axis=1)
return raw_points, occupancy_normal, label, invalid
class DG_Dataset(Dataset):
def __init__(self, config, split='train'):
self.config = config
self.split = split
self.files = {}
self.filenames = []
for ext in SPLIT_FILES[split]:
self.files[EXT_TO_NAME[ext]] = []
for sequence in SPLIT_SEQUENCES[split]:
voxels_path = os.path.join(config['GENERAL']['dataset_dir'], 'sequences', sequence, 'voxels')
if not os.path.exists(voxels_path): raise RuntimeError('Voxel directory missing: ' + voxels_path)
files = os.listdir(voxels_path)
for ext in SPLIT_FILES[split]:
comletion_data = sorted([os.path.join(voxels_path, f) for f in files if f.endswith(ext)])
if len(comletion_data) == 0: raise RuntimeError('Missing data for ' + EXT_TO_NAME[ext])
self.files[EXT_TO_NAME[ext]].extend(comletion_data)
# filename
self.filenames.extend(sorted([(sequence, os.path.splitext(f)[0]) for f in files if f.endswith('.bin')]))
self.num_files = len(self.filenames)
config['DATA_IO']['file_num'] = self.num_files
# print('Files num: %d' % self.num_files)
# sanity check:
for k, v in self.files.items():
assert (len(v) == self.num_files)
remapdict = kitti_config['learning_map']
maxkey = max(remapdict.keys())
remap_lut = np.zeros((maxkey + 100), dtype=np.int32)
remap_lut[list(remapdict.keys())] = list(remapdict.values())
remap_lut[remap_lut == 0] = 255 # map 0 to 'invalid'
remap_lut[0] = 0 # only 'empty' stays 'empty'.
self.comletion_remap_lut = remap_lut
complt_num_per_class = np.array(config['DATA_IO']['complt_num_per_class'])
compl_labelweights = complt_num_per_class / np.sum(complt_num_per_class)
self.compl_labelweights = np.power(np.amax(compl_labelweights) / compl_labelweights, 1 / 3.0)
def __len__(self):
return self.num_files
def getitem_train(self, idx):
# input, label, invalid, mask
raw_data = unpack(np.fromfile(self.files['input'][idx], dtype=np.uint8))
raw_data = raw_data.reshape(VOXEL_SCALE)
label = np.fromfile(self.files['label'][idx], dtype=np.uint16)
label = self.comletion_remap_lut[label]
label = label.reshape(VOXEL_SCALE)
invalid = unpack(np.fromfile(self.files['invalid'][idx], dtype=np.uint8))
invalid = invalid.reshape(VOXEL_SCALE)
# augmentation
raw_data_sparse, occupancy_normal, label_volume, invalid = data_augmentation(raw_data, label, invalid, self.config)
# points
point_cloud_size = occupancy_normal.shape[0]
occupancy_sparse = np.zeros([point_cloud_size,3])
occupancy_sparse = occupancy_normal[:,:3]
# on&off surface points coordinates and normals
on_surface_size = self.config['TRAIN']['G_TRAIN']['on_surface_size']
rand_idcs = np.random.choice(point_cloud_size, size=on_surface_size)
on_surface_coords = np.zeros([on_surface_size,3])
on_surface_coords = occupancy_normal[rand_idcs, :3]
on_surface_labels = label_volume[tuple(np.transpose(on_surface_coords.astype(np.int)))].reshape(-1)
on_surface_coords += 0.5
on_surface_coords = on_surface_coords / 256.0
on_surface_coords -= 0.5
on_surface_coords *= 2
on_surface_normals = occupancy_normal[rand_idcs, 3:]
on_surface_occ = np.zeros(on_surface_size)
off_count_scale = self.config['DATA_IO']['off_count_scale']
off_surface_size = on_surface_size // off_count_scale
off_surface_xy = np.random.uniform(-1, 1, size=(off_surface_size, 2))
off_surface_z = np.random.uniform(-1, -0.75, size=(off_surface_size, 1)) # -1,-0.75
off_surface_coords = np.concatenate((off_surface_xy, off_surface_z), axis=1)
off_surface_normals = np.ones((off_surface_size, 3)) * -1
if self.config['DATA_IO']['ignore_off_label']:
off_surface_labels = np.ones(off_surface_size) * 255
else:
off_surface_labels = np.ones(off_surface_size) * 0
off_surface_occ = np.ones(off_surface_size)
# off_surface vertex
if self.config['DATA_IO']['use_off_vertex']:
off_surface_v_size = on_surface_size // off_count_scale
empty = np.ones_like(label_volume)
empty[label_volume!=0] = 0
empty_vertex = np.transpose(empty.nonzero())
rand_idcs = np.random.choice(empty_vertex.shape[0], size=off_surface_v_size)
off_surface_v_coords = empty_vertex[rand_idcs, :3]
off_surface_v_coords = off_surface_v_coords * 1.0 + 0.5
off_surface_v_coords = off_surface_v_coords / 256.0
off_surface_v_coords -= 0.5
off_surface_v_coords *= 2
off_surface_v_normals = np.ones((off_surface_v_size, 3)) * -1
off_surface_v_labels = np.zeros(off_surface_v_size)
off_surface_v_occ = np.ones(off_surface_v_size)
else:
off_surface_v_size = 0
off_surface_v_coords = np.array([]).reshape(-1,3)
off_surface_v_normals = np.array([]).reshape(-1,3)
off_surface_v_labels = np.array([]).reshape(-1)
off_surface_v_occ = np.array([]).reshape(-1)
# output
coords = np.concatenate((on_surface_coords, off_surface_coords, off_surface_v_coords), axis=0)
# coords = np.concatenate((on_surface_coords, off_surface_v_coords), axis=0)
out_coords = torch.from_numpy(coords).float()
sdf = np.zeros((on_surface_size + off_surface_size + off_surface_v_size, 1)) # on-surface = 0
sdf[on_surface_size:, :] = -1 # off-surface = -1
normals = np.concatenate((on_surface_normals, off_surface_normals, off_surface_v_normals), axis=0)
labels = np.concatenate((on_surface_labels, off_surface_labels, off_surface_v_labels), axis=0)
occ = np.concatenate((on_surface_occ, off_surface_occ, off_surface_v_occ), axis=0)
out_sdf = torch.from_numpy(sdf).float()
out_normals = torch.from_numpy(normals).float()
out_labels_point = torch.from_numpy(labels).long()
out_occ = torch.from_numpy(occ).float()
out_label = torch.from_numpy(label_volume).long()
out_invalid = torch.from_numpy(invalid).long()
out_raw_data = torch.from_numpy(raw_data_sparse).float()
if self.config['TRAIN']['D_TRAIN']['D_input'] == 'occupancy':
out_raw_feat = torch.ones((len(out_raw_data), 1))
elif self.config['TRAIN']['D_TRAIN']['D_input'] == 'radial':
out_raw_feat = torch.norm(out_raw_data-torch.tensor([0,127.5,0]), p=2, dim=1).reshape(-1,1)
elif self.config['TRAIN']['D_TRAIN']['D_input'] == 'radial_height':
out_raw_feat = torch.norm(out_raw_data-torch.tensor([0,127.5,0]), p=2, dim=1).reshape(-1,1)
z = out_raw_data[:,2].reshape(-1,1)
out_raw_feat = torch.cat((out_raw_feat, z), dim=1)
out_occupancy = torch.from_numpy(occupancy_sparse).float()
return idx, \
{'coords': out_coords}, \
{'sdf': out_sdf, 'normals': out_normals, 'label': out_label, 'invalid': out_invalid, 'label_points': out_labels_point, 'occ': out_occ}, \
{'raw': out_raw_data, 'raw_feat': out_raw_feat, 'occupancy': out_occupancy}
def getitem_valid(self, idx):
# input, label, invalid, mask
raw_data = unpack(np.fromfile(self.files['input'][idx], dtype=np.uint8))
raw_data = raw_data.reshape(VOXEL_SCALE)
raw_pos = raw_data.nonzero()
raw_data_sparse = np.transpose(raw_pos)
label = np.fromfile(self.files['label'][idx], dtype=np.uint16)
label = self.comletion_remap_lut[label]
label = label.reshape(VOXEL_SCALE)
invalid = unpack(np.fromfile(self.files['invalid'][idx], dtype=np.uint8))
invalid = invalid.reshape(VOXEL_SCALE)
mask = get_eval_mask(label, invalid)
out_raw_data = torch.from_numpy(raw_data_sparse).float()
if self.config['TRAIN']['D_TRAIN']['D_input'] == 'occupancy':
out_raw_feat = torch.ones((len(out_raw_data), 1))
elif self.config['TRAIN']['D_TRAIN']['D_input'] == 'radial':
out_raw_feat = torch.norm(out_raw_data-torch.tensor([0,127.5,0]), p=2, dim=1).reshape(-1,1)
elif self.config['TRAIN']['D_TRAIN']['D_input'] == 'radial_height':
out_raw_feat = torch.norm(out_raw_data-torch.tensor([0,127.5,0]), p=2, dim=1).reshape(-1,1)
z = out_raw_data[:,2].reshape(-1,1)
out_raw_feat = torch.cat((out_raw_feat, z), dim=1)
index_info = self.filenames[idx][0]+'_'+self.filenames[idx][1]
return index_info, \
{'raw': out_raw_data, 'raw_feat': out_raw_feat}, \
{'label': label, 'mask': mask}
def __getitem__(self, idx):
if self.split == 'train':
return self.getitem_train(idx)
elif self.split == 'valid':
return self.getitem_valid(idx)
def DG_DataMerge_train(batch):
out_coords = []
out_sdf = []
out_normals = []
out_labels = []
out_invalid = []
out_label_points = []
out_occ = []
out_indices = []
out_raw_data = []
out_feature = []
out_occupancy = []
for num, example in enumerate(batch):
idx, points, gt, raw_occupancy = example
out_indices.append(idx)
out_coords.append(points['coords'])
out_sdf.append(gt['sdf'])
out_normals.append(gt['normals'])
out_labels.append(gt['label'])
out_invalid.append(gt['invalid'])
out_label_points.append(gt['label_points'])
out_occ.append(gt['occ'])
out_raw_data.append(raw_occupancy['raw'])
out_feature.append(raw_occupancy['raw_feat'])
out_occupancy.append(raw_occupancy['occupancy'])
out_coords = torch.stack(out_coords)
out_sdf = torch.stack(out_sdf)
out_normals = torch.stack(out_normals)
out_labels = torch.stack(out_labels)
out_invalid = torch.stack(out_invalid)
out_label_points = torch.stack(out_label_points)
out_occ = torch.stack(out_occ)
indices_out = out_indices
points_out = {'coords': out_coords}
gt_output = {'sdf': out_sdf, 'normals': out_normals, 'label': out_labels, 'invalid': out_invalid, 'out_label_points': out_label_points, 'occ': out_occ}
raw_data_out = ME.utils.batched_coordinates(out_raw_data)
feature_out = torch.cat(out_feature)
occupancy_out = ME.utils.batched_coordinates(out_occupancy)
return indices_out, points_out, gt_output, raw_data_out, feature_out, occupancy_out
def DG_DataMerge_valid(batch):
out_labels = []
out_masks = []
out_indices = []
out_raw_data = []
out_feature = []
out_occupancy = []
for num, example in enumerate(batch):
idx, raw_occupancy, eval_info = example
out_indices.append(idx)
out_labels.append(eval_info['label'])
out_masks.append(eval_info['mask'])
out_raw_data.append(raw_occupancy['raw'])
out_feature.append(raw_occupancy['raw_feat'])
out_occupancy.append(torch.zeros([1,3])) # no use
indices_out = out_indices
out_raw = np.stack(out_raw_data, axis=0)
out_labels = np.stack(out_labels, axis=0)
out_masks = np.stack(out_masks, axis=0)
eval_out = {'raw': out_raw,'label': out_labels, 'mask': out_masks}
raw_data_out = ME.utils.batched_coordinates(out_raw_data)
feature_out = torch.cat(out_feature)
occupancy_out = ME.utils.batched_coordinates(out_occupancy)
return indices_out, eval_out, raw_data_out, feature_out, occupancy_out