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
|
# ----------------------------------------------------------------------------
# - Open3D: www.open3d.org -
# ----------------------------------------------------------------------------
# The MIT License (MIT)
#
# Copyright (c) 2018-2021 www.open3d.org
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in
# all copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
# IN THE SOFTWARE.
# ----------------------------------------------------------------------------
import open3d as o3d
import numpy as np
from scipy.spatial import cKDTree
import pytest
import mltest
import torch
import tensorflow as tf
# skip all tests if the ml ops were not built
pytestmark = mltest.default_marks
# the supported dtypes for the point coordinates
dtypes = pytest.mark.parametrize('dtype', [np.float32, np.float64])
@dtypes
@mltest.parametrize.ml_cpu_only
@pytest.mark.parametrize('num_points_queries', [(2, 5), (31, 33), (33, 31),
(123, 345)])
@pytest.mark.parametrize('metric', ['L1', 'L2'])
@pytest.mark.parametrize('ignore_query_point', [False, True])
@pytest.mark.parametrize('return_distances', [False, True])
@pytest.mark.parametrize('index_dtype', ['int32', 'int64'])
def test_knn_search(dtype, ml, num_points_queries, metric, ignore_query_point,
return_distances, index_dtype):
rng = np.random.RandomState(123)
num_points, num_queries = num_points_queries
points = rng.random(size=(num_points, 3)).astype(dtype)
if ignore_query_point:
queries = points
else:
queries = rng.random(size=(num_queries, 3)).astype(dtype)
k = rng.randint(1, 11)
# kd tree for computing the ground truth
tree = cKDTree(points, copy_data=True)
p_norm = {'L1': 1, 'L2': 2, 'Linf': np.inf}[metric]
if k > num_points:
gt_neighbors_index = [tree.query(q, k, p=p_norm) for q in queries]
gt_neighbors_index = [
idxs[np.isfinite(dists)] for dists, idxs in gt_neighbors_index
]
else:
gt_neighbors_index = [tree.query(q, k, p=p_norm)[1] for q in queries]
if ml.module.__name__ == 'tensorflow':
index_dtype_ = {'int32': tf.int32, 'int64': tf.int64}[index_dtype]
elif ml.module.__name__ == 'torch':
index_dtype_ = {'int32': torch.int32, 'int64': torch.int64}[index_dtype]
else:
raise Exception('Unsupported ml framework')
layer = ml.layers.KNNSearch(metric=metric,
ignore_query_point=ignore_query_point,
return_distances=return_distances,
index_dtype=index_dtype_)
ans = mltest.run_op(
ml,
ml.device,
True,
layer,
points,
queries=queries,
k=k,
)
index_dtype_np = {'int32': np.int32, 'int64': np.int64}[index_dtype]
assert ans.neighbors_index.dtype == index_dtype_np
for i, q in enumerate(queries):
# check neighbors
start = ans.neighbors_row_splits[i]
end = ans.neighbors_row_splits[i + 1]
q_neighbors_index = ans.neighbors_index[start:end]
if k == 1:
gt_set = set([gt_neighbors_index[i]])
else:
gt_set = set(gt_neighbors_index[i])
if ignore_query_point:
gt_set.remove(i)
assert gt_set == set(q_neighbors_index)
# check distances
if return_distances:
q_neighbors_dist = ans.neighbors_distance[start:end]
for j, dist in zip(q_neighbors_index, q_neighbors_dist):
if metric == 'L2':
gt_dist = np.sum((q - points[j])**2)
else:
gt_dist = np.linalg.norm(q - points[j], ord=p_norm)
np.testing.assert_allclose(dist, gt_dist, rtol=1e-7, atol=1e-8)
@mltest.parametrize.ml_cpu_only
def test_knn_search_empty_point_sets(ml):
rng = np.random.RandomState(123)
dtype = np.float32
# no query points
points = rng.random(size=(100, 3)).astype(dtype)
queries = rng.random(size=(0, 3)).astype(dtype)
k = rng.randint(1, 11)
layer = ml.layers.KNNSearch(return_distances=True)
ans = mltest.run_op(
ml,
ml.device,
True,
layer,
points,
queries=queries,
k=k,
)
assert ans.neighbors_index.shape == (0,)
assert ans.neighbors_row_splits.shape == (1,)
assert ans.neighbors_distance.shape == (0,)
# no input points
points = rng.random(size=(0, 3)).astype(dtype)
queries = rng.random(size=(100, 3)).astype(dtype)
layer = ml.layers.KNNSearch(return_distances=True)
ans = mltest.run_op(
ml,
ml.device,
True,
layer,
points,
queries=queries,
k=k,
)
assert ans.neighbors_index.shape == (0,)
assert ans.neighbors_row_splits.shape == (101,)
np.testing.assert_array_equal(np.zeros_like(ans.neighbors_row_splits),
ans.neighbors_row_splits)
assert ans.neighbors_distance.shape == (0,)
@mltest.parametrize.ml_cpu_only
@pytest.mark.parametrize('batch_size', [2, 3, 8])
def test_knn_search_batches(ml, batch_size):
dtype = np.float32
metric = 'L2'
p_norm = {'L1': 1, 'L2': 2, 'Linf': np.inf}[metric]
ignore_query_point = False
return_distances = True
rng = np.random.RandomState(123)
# create array defining start and end of each batch
points_row_splits = np.zeros(shape=(batch_size + 1,), dtype=np.int64)
queries_row_splits = np.zeros(shape=(batch_size + 1,), dtype=np.int64)
for i in range(batch_size):
points_row_splits[i + 1] = rng.randint(15) + points_row_splits[i]
queries_row_splits[i + 1] = rng.randint(15) + queries_row_splits[i]
num_points = points_row_splits[-1]
num_queries = queries_row_splits[-1]
points = rng.random(size=(num_points, 3)).astype(dtype)
if ignore_query_point:
queries = points
queries_row_splits = points_row_splits
else:
queries = rng.random(size=(num_queries, 3)).astype(dtype)
k = rng.randint(1, 11)
# kd tree for computing the ground truth
gt_neighbors_index = []
for i in range(batch_size):
points_i = points[points_row_splits[i]:points_row_splits[i + 1]]
queries_i = queries[queries_row_splits[i]:queries_row_splits[i + 1]]
tree = cKDTree(points_i, copy_data=True)
if k > points_i.shape[0]:
tmp = [tree.query(q, k, p=p_norm) for q in queries_i]
tmp = [
list(idxs[np.isfinite(dists)] + points_row_splits[i])
for dists, idxs in tmp
]
else:
tmp = [
list(tree.query(q, k, p=p_norm)[1] + points_row_splits[i])
for q in queries_i
]
gt_neighbors_index.extend(tmp)
layer = ml.layers.KNNSearch(metric=metric,
ignore_query_point=ignore_query_point,
return_distances=return_distances)
ans = mltest.run_op(ml,
ml.device,
True,
layer,
points,
queries=queries,
k=k,
points_row_splits=points_row_splits,
queries_row_splits=queries_row_splits)
for i, q in enumerate(queries):
# check neighbors
start = ans.neighbors_row_splits[i]
end = ans.neighbors_row_splits[i + 1]
q_neighbors_index = ans.neighbors_index[start:end]
if k == 1:
gt_set = set([gt_neighbors_index[i]])
else:
gt_set = set(gt_neighbors_index[i])
if ignore_query_point:
gt_set.remove(i)
assert gt_set == set(q_neighbors_index)
# check distances
if return_distances:
q_neighbors_dist = ans.neighbors_distance[start:end]
for j, dist in zip(q_neighbors_index, q_neighbors_dist):
if metric == 'L2':
gt_dist = np.sum((q - points[j])**2)
else:
gt_dist = np.linalg.norm(q - points[j], ord=p_norm)
np.testing.assert_allclose(dist, gt_dist, rtol=1e-7, atol=1e-8)
|
