-
Notifications
You must be signed in to change notification settings - Fork 1
/
dataset.py
160 lines (130 loc) · 5.75 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
from datetime import datetime
from pathlib import Path
from typing import Sequence, Tuple, Union, Optional, List
import numpy as np
def broaden_annotation(point_cloud: np.ndarray,
annotation: np.ndarray,
radius: float = 0.01) -> np.ndarray:
output = []
annotation_cloud = point_cloud[annotation.astype(np.bool)]
for annotation_point in annotation_cloud:
ds = np.abs(np.linalg.norm(annotation_point - point_cloud, axis=1))
output.append(ds < radius)
return np.logical_or.reduce(output).astype(np.uint8)
class Dataset(Sequence):
def __init__(self, root_path: Path, only_annotated: bool = True,
selection: Optional[List[int]] = None,
broaden_annotations: bool = False):
self._root_path = root_path
self._only_annotated = only_annotated
self._selection = selection
self._broaden_annotations = broaden_annotations
def __len__(self):
if self._selection is not None:
return len(self._selection)
if self._only_annotated:
return len([a for a in self._root_path.glob("*_annotation*") if a.is_file()])
return len([a for a in self._root_path.glob("*_data*") if a.is_file()])
def _get_item_index(self, index: int) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
data = sorted(self._root_path.glob("*_data*"))
while True:
if self._selection is not None:
selected_sample_path = data[self._selection[index]]
else:
selected_sample_path = data[index]
selected_sample = selected_sample_path.name.split("_data")[0]
try:
return self._get_item_str(selected_sample)
except Exception as e:
if str(e) != "No annotation":
raise
index += 1
def _get_item_datetime(
self, timestamp: datetime
) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
return self._get_item_str(Dataset.timestamp(timestamp))
def _get_item_str(
self, index: str
) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
item_path = self._root_path / (index + "_data.npy")
if not item_path.exists():
raise Exception(f"index {index} doesn't exist in dataset.")
point_cloud = np.load(str(item_path))
annotation_path = self._root_path / (index + "_annotation.npy")
if annotation_path.exists():
annotation_cloud = np.load(str(annotation_path))
if self._broaden_annotations:
annotation_cloud = broaden_annotation(point_cloud, annotation_cloud)
else:
if self._only_annotated:
raise Exception("No annotation")
annotation_cloud = np.zeros([point_cloud.shape[0],], dtype=np.uint8)
return point_cloud, np.zeros((point_cloud.shape[0], 0)), annotation_cloud
def __getitem__(
self, index: Union[int, datetime]
) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
if isinstance(index, datetime):
return self._get_item_datetime(index)
elif isinstance(index, int):
return self._get_item_index(index)
else:
raise Exception("invalid key!")
def __setitem__(self, index: datetime, value: np.ndarray) -> None:
root = self._root_path / Dataset.timestamp(index)
self._root_path.mkdir(parents=True, exist_ok=True)
np.save(str(root) + "_data", value)
def set_annotation(self, index: datetime, value: np.ndarray) -> None:
root = self._root_path / Dataset.timestamp(index)
self._root_path.mkdir(parents=True, exist_ok=True)
np.save(str(root) + "_annotation", value)
@classmethod
def timestamp(cls, time: Optional[datetime]) -> str:
input_datetime: datetime = datetime.now()
if time is not None:
input_datetime = time
return "%04.i_%02.i_%02.i__%02.i_%02.i_%02.i_%06.i000" % (
input_datetime.year,
input_datetime.month,
input_datetime.day,
input_datetime.hour,
input_datetime.minute,
input_datetime.second,
input_datetime.microsecond,
)
def split(self, percentage: float = 0.8) -> "Tuple[Dataset, Dataset]":
indices = list(range(len(self)))
np.random.seed(3)
np.random.shuffle(indices)
split_index = int(percentage*len(indices))
return Dataset(
self._root_path, self._only_annotated, selection=indices[: split_index]
), Dataset(
self._root_path, self._only_annotated, selection=indices[split_index:]
)
class DatasetMerged(Sequence):
def __init__(self, datasets: List[Dataset],
selection: Optional[List[int]] = None):
self._datasets = datasets
self._selection = selection
def __len__(self):
if self._selection is not None:
return len(self._selection)
return sum(map(lambda s: len(s), self._datasets))
def __getitem__(self, item: int) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
if self._selection is not None:
item = self._selection[item]
for dataset in self._datasets:
if item >= len(dataset):
item -= len(dataset)
else:
return dataset[item]
def split(self, percentage: float = 0.8) -> "Tuple[DatasetMerged, DatasetMerged]":
indices = list(range(len(self)))
np.random.seed(3)
np.random.shuffle(indices)
split_index = int(percentage * len(indices))
return DatasetMerged(
self._datasets, selection=indices[: split_index]
), DatasetMerged(
self._datasets, selection=indices[split_index:]
)