forked from facebookresearch/SparseConvNet
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathioLayers.py
More file actions
294 lines (250 loc) · 8.45 KB
/
Copy pathioLayers.py
File metadata and controls
294 lines (250 loc) · 8.45 KB
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
# Copyright 2016-present, Facebook, Inc.
# All rights reserved.
#
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.
import sparseconvnet.SCN
from torch.autograd import Function
from torch.nn import Module, Parameter
from .utils import *
from .sparseConvNetTensor import SparseConvNetTensor
from .metadata import Metadata
class InputLayer(Module):
"""
Takes a tuple (coords, features, batch_size [optional])
* coords is 2d torch.LongTensor with size
N x dimension (batch size == 1)
or
N x (dimension+1) (first d columns are coordinates, last column is batch index)
* features is a CPU or CUDA float tensor with size
N x n_feature_planes
* batch_size if given, set a lower bound on the the number of samples in the output tensor.
Batch size can normally be inferred from the last column of coords, but this may fail if
some of the batch items are totally empty.
In case of repetition in coords:
mode == 0 if the input is guaranteed to have no duplicates
mode == 1 to use the last item at each spatial location
mode == 2 to keep the first item at each spatial location
mode == 3 to sum feature vectors sharing one spatial location
mode == 4 to average feature vectors at each spatial location
Output is a SparseConvNetTensor
"""
def __init__(self, dimension, spatial_size, mode=3):
Module.__init__(self)
self.dimension = dimension
self.spatial_size = toLongTensor(dimension, spatial_size)
self.mode = mode
self.device = None
def to(self, device):
self.device=device
return self
def forward(self, input):
output = SparseConvNetTensor(
metadata=Metadata(
self.dimension),
spatial_size=self.spatial_size)
output.features = InputLayerFunction.apply(
self.dimension,
output.metadata,
self.spatial_size,
input[0].cpu().long(),
input[1].to(self.device) if self.device else input[1],
0 if len(input) == 2 else input[2],
self.mode
)
return output
class OutputLayer(Module):
"""
Used in conjunction with an InputLayer for 'autoencoder' style networks
Takes a SparseConvNetTensor and results a float Tensor of size
N x n_feature_planes
N is defined by the InputLayer
Behavior during forward-/back-propagation depends on the InputLayer's mode
"""
def __init__(self, dimension):
Module.__init__(self)
self.dimension = dimension
def forward(self, input):
output = OutputLayerFunction.apply(
self.dimension,
input.metadata,
input.features
)
return output
class BLInputLayer(Module):
"""
Takes a tuple (coords, features)
* coords is 3d torch.LongTensor with size
batch_size x length x dimension
Coordinates should be >=0, or -1 to indicate 'empty'
* features is a 3d CPU or CUDA float Tensor with size
batch_size x length x n_feature_planes
mode == 0 Assumes that for each coords[i, :], the locations are unique and not 'empty'.
mode == 1 Use the last item at each spatial location
mode == 2 Keep the first item at each spatial location
mode == 3 Sum feature vectors sharing one spatial location
mode == 4 Average feature vectors at each spatial location
Output is a SparseConvNetTensor
"""
def __init__(self, dimension, spatial_size, mode=3):
Module.__init__(self)
self.dimension = dimension
self.spatial_size = toLongTensor(dimension, spatial_size)
self.mode = mode
self.device = None
def to(self, device):
self.device=device
return self
def forward(self, input):
output = SparseConvNetTensor(
metadata=Metadata(
self.dimension),
spatial_size=self.spatial_size)
output.features = BLInputLayerFunction.apply(
self.dimension,
output.metadata,
self.spatial_size,
input[0].cpu().long(),
input[1].to(self.device) if self.device else input[1],
self.mode
)
return output
class BLOutputLayer(Module):
"""
Used in conjunction with a BLInputLayer for 'autoencoder' style networks
Takes a SparseConvNetTensor and results a float Tensor of batch_size
batch_size x length x n_feature_planes
batch_size and length are defined by the BLInputLayer
Behavior during forward-/back-propagation depends on the BLInputLayer's mode
"""
def __init__(self, dimension):
Module.__init__(self)
self.dimension = dimension
def forward(self, input):
output = BLOutputLayerFunction.apply(
self.dimension,
input.metadata,
input.features
)
return output
class InputLayerFunction(Function):
@staticmethod
def forward(
ctx,
dimension,
metadata,
spatial_size,
coords,
input_features,
batch_size,
mode):
output_features = input_features.new()
ctx.dimension = dimension
ctx.metadata_ = metadata
sparseconvnet.SCN.InputLayer_updateOutput(
metadata,
spatial_size,
coords,
input_features.contiguous(),
output_features,
batch_size,
mode
)
return output_features
@staticmethod
def backward(ctx, grad_output):
grad_input = grad_output.new()
sparseconvnet.SCN.InputLayer_updateGradInput(
ctx.metadata_,
grad_input,
grad_output.contiguous())
return None, None, None, None, grad_input, None, None
class OutputLayerFunction(Function):
@staticmethod
def forward(
ctx,
dimension,
metadata,
input_features):
output_features = input_features.new()
ctx.metadata_ = metadata
ctx.dimension = dimension
sparseconvnet.SCN.OutputLayer_updateOutput(
metadata,
input_features.contiguous(),
output_features
)
return output_features
@staticmethod
def backward(ctx, grad_output):
grad_input = grad_output.new()
grad_output=grad_output.contiguous()
sparseconvnet.SCN.OutputLayer_updateGradInput(
ctx.metadata_,
grad_input,
grad_output.contiguous())
return None, None, grad_input
class BLInputLayerFunction(Function):
@staticmethod
def forward(
ctx,
dimension,
metadata,
spatial_size,
coords,
input_features,
mode):
output_features = input_features.new()
ctx.metadata_ = metadata
ctx.dimension = dimension
sparseconvnet.SCN.BLInputLayer_updateOutput(
metadata,
spatial_size,
coords,
input_features.contiguous(),
output_features,
mode
)
return output_features
@staticmethod
def backward(ctx, grad_output):
grad_input = grad_output.new()
sparseconvnet.SCN.BLInputLayer_updateGradInput(
ctx.metadata_,
grad_input,
grad_output.contiguous())
return None, None, None, None, grad_input, None
class BLOutputLayerFunction(Function):
@staticmethod
def forward(
ctx,
dimension,
metadata,
input_features):
output_features = input_features.new()
ctx.metadata_ = metadata
ctx.dimension = dimension
sparseconvnet.SCN.BLOutputLayer_updateOutput(
metadata,
input_features.contiguous(),
output_features
)
return output_features
@staticmethod
def backward(ctx, grad_output):
grad_input = grad_output.new()
sparseconvnet.SCN.BLOutputLayer_updateGradInput(
ctx.metadata_,
grad_input,
grad_output.contiguous())
return None, None, grad_input
class InputLayerInput(object):
def __init__(self,coords,features):
self.x=[coords,features]
def __getitem__(self,n):
return self.x[n]
def __len__(self):
return 2
def cuda(self):
self.x[1]=self.x[1].cuda()
return self