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
390
|
// Copyright 2016, Tobias Hermann.
// https://github.com/Dobiasd/frugally-deep
// Distributed under the MIT License.
// (See accompanying LICENSE file or at
// https://opensource.org/licenses/MIT)
#pragma once
#include "fdeep/common.hpp"
#include "fdeep/shape2.hpp"
#include "fdeep/tensor_shape_variable.hpp"
#include <algorithm>
#include <cstddef>
#include <cstdlib>
#include <string>
#include <vector>
namespace fdeep {
namespace internal {
class tensor_shape {
public:
// The outer (left-most) dimensions are not used for batch prediction.
// If you like to do multiple forward passes on a model at once,
// use fdeep::model::predict_multi instead.
explicit tensor_shape(
std::size_t size_dim_5,
std::size_t size_dim_4,
std::size_t height,
std::size_t width,
std::size_t depth)
: size_dim_5_(size_dim_5)
, size_dim_4_(size_dim_4)
, height_(height)
, width_(width)
, depth_(depth)
, rank_(5)
{
}
explicit tensor_shape(
std::size_t size_dim_4,
std::size_t height,
std::size_t width,
std::size_t depth)
: size_dim_5_(1)
, size_dim_4_(size_dim_4)
, height_(height)
, width_(width)
, depth_(depth)
, rank_(4)
{
}
explicit tensor_shape(
std::size_t height,
std::size_t width,
std::size_t depth)
: size_dim_5_(1)
, size_dim_4_(1)
, height_(height)
, width_(width)
, depth_(depth)
, rank_(3)
{
}
explicit tensor_shape(
std::size_t width,
std::size_t depth)
: size_dim_5_(1)
, size_dim_4_(1)
, height_(1)
, width_(width)
, depth_(depth)
, rank_(2)
{
}
explicit tensor_shape(
std::size_t depth)
: size_dim_5_(1)
, size_dim_4_(1)
, height_(1)
, width_(1)
, depth_(depth)
, rank_(1)
{
}
std::size_t volume() const
{
return size_dim_5_ * size_dim_4_ * height_ * width_ * depth_;
}
void assert_is_shape_2() const
{
assertion(
size_dim_5_ == 1 && size_dim_4_ == 1 && depth_ == 1,
"Only height and width may be not equal 1.");
}
void assert_is_shape_3() const
{
assertion(
size_dim_5_ == 1 && size_dim_4_ == 1,
"Only height, width and depth may be not equal 1.");
}
shape2 without_depth() const
{
assert_is_shape_3();
return shape2(height_, width_);
}
std::size_t rank() const
{
assertion(rank_ >= 1 && rank_ <= 5, "Invalid rank");
return rank_;
}
std::size_t minimal_rank() const
{
if (size_dim_5_ > 1)
return 5;
if (size_dim_4_ > 1)
return 4;
if (height_ > 1)
return 3;
if (width_ > 1)
return 2;
return 1;
}
void shrink_rank()
{
rank_ = minimal_rank();
}
void shrink_rank_with_min(std::size_t min_rank_to_keep)
{
rank_ = fplus::max(minimal_rank(), min_rank_to_keep);
}
void maximize_rank()
{
rank_ = 5;
}
std::vector<std::size_t> dimensions() const
{
if (rank() == 5)
return { size_dim_5_, size_dim_4_, height_, width_, depth_ };
if (rank() == 4)
return { size_dim_4_, height_, width_, depth_ };
if (rank() == 3)
return { height_, width_, depth_ };
if (rank() == 2)
return { width_, depth_ };
return { depth_ };
}
std::size_t size_dim_5_;
std::size_t size_dim_4_;
std::size_t height_;
std::size_t width_;
std::size_t depth_;
private:
std::size_t rank_;
};
inline tensor_shape create_tensor_shape_from_dims(
const std::vector<std::size_t>& dimensions)
{
assertion(dimensions.size() >= 1 && dimensions.size() <= 5,
"Invalid tensor-shape dimensions");
if (dimensions.size() == 5)
return tensor_shape(
dimensions[0],
dimensions[1],
dimensions[2],
dimensions[3],
dimensions[4]);
if (dimensions.size() == 4)
return tensor_shape(
dimensions[0],
dimensions[1],
dimensions[2],
dimensions[3]);
if (dimensions.size() == 3)
return tensor_shape(
dimensions[0],
dimensions[1],
dimensions[2]);
if (dimensions.size() == 2)
return tensor_shape(
dimensions[0],
dimensions[1]);
return tensor_shape(dimensions[0]);
}
inline tensor_shape make_tensor_shape_with(
const tensor_shape& default_shape,
const tensor_shape_variable shape)
{
if (shape.rank() == 1)
return tensor_shape(
fplus::just_with_default(default_shape.depth_, shape.depth_));
if (shape.rank() == 2)
return tensor_shape(
fplus::just_with_default(default_shape.width_, shape.width_),
fplus::just_with_default(default_shape.depth_, shape.depth_));
if (shape.rank() == 3)
return tensor_shape(
fplus::just_with_default(default_shape.height_, shape.height_),
fplus::just_with_default(default_shape.width_, shape.width_),
fplus::just_with_default(default_shape.depth_, shape.depth_));
if (shape.rank() == 4)
return tensor_shape(
fplus::just_with_default(default_shape.size_dim_4_, shape.size_dim_4_),
fplus::just_with_default(default_shape.height_, shape.height_),
fplus::just_with_default(default_shape.width_, shape.width_),
fplus::just_with_default(default_shape.depth_, shape.depth_));
else
return tensor_shape(
fplus::just_with_default(default_shape.size_dim_5_, shape.size_dim_5_),
fplus::just_with_default(default_shape.size_dim_4_, shape.size_dim_4_),
fplus::just_with_default(default_shape.height_, shape.height_),
fplus::just_with_default(default_shape.width_, shape.width_),
fplus::just_with_default(default_shape.depth_, shape.depth_));
}
inline tensor_shape derive_fixed_tensor_shape(
std::size_t values,
const tensor_shape_variable shape)
{
const auto inferred = values / shape.minimal_volume();
return make_tensor_shape_with(
tensor_shape(inferred, inferred, inferred, inferred, inferred), shape);
}
inline bool tensor_shape_equals_tensor_shape_variable(
const tensor_shape& lhs, const tensor_shape_variable& rhs)
{
return (rhs.rank() == lhs.rank()) && (rhs.size_dim_5_.is_nothing() || lhs.size_dim_5_ == rhs.size_dim_5_.unsafe_get_just()) && (rhs.size_dim_4_.is_nothing() || lhs.size_dim_4_ == rhs.size_dim_4_.unsafe_get_just()) && (rhs.height_.is_nothing() || lhs.height_ == rhs.height_.unsafe_get_just()) && (rhs.width_.is_nothing() || lhs.width_ == rhs.width_.unsafe_get_just()) && (rhs.depth_.is_nothing() || lhs.depth_ == rhs.depth_.unsafe_get_just());
}
inline bool operator==(const tensor_shape& lhs, const tensor_shape_variable& rhs)
{
return tensor_shape_equals_tensor_shape_variable(lhs, rhs);
}
inline bool operator==(const std::vector<tensor_shape>& lhss,
const std::vector<tensor_shape_variable>& rhss)
{
return fplus::all(fplus::zip_with(tensor_shape_equals_tensor_shape_variable,
lhss, rhss));
}
inline bool operator==(const tensor_shape& lhs, const tensor_shape& rhs)
{
return lhs.rank() == rhs.rank() && lhs.size_dim_5_ == rhs.size_dim_5_ && lhs.size_dim_4_ == rhs.size_dim_4_ && lhs.height_ == rhs.height_ && lhs.width_ == rhs.width_ && lhs.depth_ == rhs.depth_;
}
inline tensor_shape tensor_shape_with_changed_rank(const tensor_shape& s, std::size_t rank)
{
assertion(rank >= 1 && rank <= 5, "Invalid target rank");
if (rank == 4) {
assertion(s.size_dim_5_ == 1, "Invalid target rank");
return tensor_shape(s.size_dim_4_, s.height_, s.width_, s.depth_);
}
if (rank == 3) {
assertion(s.size_dim_5_ == 1, "Invalid target rank");
assertion(s.size_dim_4_ == 1, "Invalid target rank");
return tensor_shape(s.height_, s.width_, s.depth_);
}
if (rank == 2) {
assertion(s.size_dim_5_ == 1, "Invalid target rank");
assertion(s.size_dim_4_ == 1, "Invalid target rank");
assertion(s.height_ == 1, "Invalid target rank");
return tensor_shape(s.width_, s.depth_);
}
if (rank == 1) {
assertion(s.size_dim_5_ == 1, "Invalid target rank");
assertion(s.size_dim_4_ == 1, "Invalid target rank");
assertion(s.height_ == 1, "Invalid target rank");
assertion(s.width_ == 1, "Invalid target rank");
return tensor_shape(s.depth_);
}
return tensor_shape(s.size_dim_5_, s.size_dim_4_, s.height_, s.width_, s.depth_);
}
inline tensor_shape dilate_tensor_shape(
const shape2& dilation_rate, const tensor_shape& s)
{
assertion(dilation_rate.height_ >= 1, "invalid dilation rate");
assertion(dilation_rate.width_ >= 1, "invalid dilation rate");
const std::size_t height = s.height_ + (s.height_ - 1) * (dilation_rate.height_ - 1);
const std::size_t width = s.width_ + (s.width_ - 1) * (dilation_rate.width_ - 1);
return tensor_shape_with_changed_rank(
tensor_shape(s.size_dim_5_, s.size_dim_4_, height, width, s.depth_),
s.rank());
}
inline std::size_t get_tensor_shape_dimension_by_index(const tensor_shape& s,
const std::size_t idx)
{
if (idx == 0)
return s.size_dim_5_;
if (idx == 1)
return s.size_dim_4_;
if (idx == 2)
return s.height_;
if (idx == 3)
return s.width_;
if (idx == 4)
return s.depth_;
raise_error("Invalid tensor_shape index.");
return 0;
}
inline tensor_shape change_tensor_shape_dimension_by_index(const tensor_shape& in,
const std::size_t idx, const std::size_t dim)
{
assertion(idx <= 4, "Invalid dimension index");
assertion(dim > 0, "Invalid dimension size");
const std::size_t out_rank = std::max<std::size_t>(5 - idx, in.rank());
assertion(out_rank >= 1 && out_rank <= 5, "Invalid target rank");
const std::size_t size_dim_5 = idx == 0 ? dim : in.size_dim_5_;
const std::size_t size_dim_4 = idx == 1 ? dim : in.size_dim_4_;
const std::size_t height = idx == 2 ? dim : in.height_;
const std::size_t width = idx == 3 ? dim : in.width_;
const std::size_t depth = idx == 4 ? dim : in.depth_;
if (out_rank == 1)
return tensor_shape(depth);
if (out_rank == 2)
return tensor_shape(width, depth);
if (out_rank == 3)
return tensor_shape(height, width, depth);
if (out_rank == 4)
return tensor_shape(size_dim_4, height, width, depth);
return tensor_shape(size_dim_5, size_dim_4, height, width, depth);
}
}
using tensor_shape = internal::tensor_shape;
inline std::string show_tensor_shape(const tensor_shape& s)
{
const std::vector<std::size_t> dimensions = {
s.size_dim_5_,
s.size_dim_4_,
s.height_,
s.width_,
s.depth_
};
return std::to_string(s.rank()) + fplus::show_cont_with_frame(", ", "(", ")", fplus::drop(5 - s.rank(), dimensions));
}
inline std::string show_tensor_shapes(
const std::vector<tensor_shape>& shapes)
{
return fplus::show_cont(fplus::transform(show_tensor_shape, shapes));
}
template <typename F>
void loop_over_all_dims(const tensor_shape& shape, F f)
{
for (std::size_t dim5 = 0; dim5 < shape.size_dim_5_; ++dim5) {
for (std::size_t dim4 = 0; dim4 < shape.size_dim_4_; ++dim4) {
for (std::size_t y = 0; y < shape.height_; ++y) {
for (std::size_t x = 0; x < shape.width_; ++x) {
for (std::size_t z = 0; z < shape.depth_; ++z) {
f(dim5, dim4, y, x, z);
}
}
}
}
}
}
}
|
