tesseract  5.0.0
input.h
Go to the documentation of this file.
1 // File: input.h
3 // Description: Input layer class for neural network implementations.
4 // Author: Ray Smith
5 //
6 // (C) Copyright 2014, Google Inc.
7 // Licensed under the Apache License, Version 2.0 (the "License");
8 // you may not use this file except in compliance with the License.
9 // You may obtain a copy of the License at
10 // http://www.apache.org/licenses/LICENSE-2.0
11 // Unless required by applicable law or agreed to in writing, software
12 // distributed under the License is distributed on an "AS IS" BASIS,
13 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 // See the License for the specific language governing permissions and
15 // limitations under the License.
17 
18 #ifndef TESSERACT_LSTM_INPUT_H_
19 #define TESSERACT_LSTM_INPUT_H_
20 
21 #include "network.h"
22 
23 namespace tesseract {
24 
25 class ScrollView;
26 
27 class Input : public Network {
28 public:
29  TESS_API
30  Input(const std::string &name, int ni, int no);
31  TESS_API
32  Input(const std::string &name, const StaticShape &shape);
33  ~Input() override = default;
34 
35  std::string spec() const override {
36  return std::to_string(shape_.batch()) + "," +
37  std::to_string(shape_.height()) + "," +
38  std::to_string(shape_.width()) + "," +
39  std::to_string(shape_.depth());
40  }
41 
42  // Returns the required shape input to the network.
43  StaticShape InputShape() const override {
44  return shape_;
45  }
46  // Returns the shape output from the network given an input shape (which may
47  // be partially unknown ie zero).
49  [[maybe_unused]] const StaticShape &input_shape) const override {
50  return shape_;
51  }
52  // Writes to the given file. Returns false in case of error.
53  // Should be overridden by subclasses, but called by their Serialize.
54  bool Serialize(TFile *fp) const override;
55  // Reads from the given file. Returns false in case of error.
56  bool DeSerialize(TFile *fp) override;
57 
58  // Returns an integer reduction factor that the network applies to the
59  // time sequence. Assumes that any 2-d is already eliminated. Used for
60  // scaling bounding boxes of truth data.
61  // WARNING: if GlobalMinimax is used to vary the scale, this will return
62  // the last used scale factor. Call it before any forward, and it will return
63  // the minimum scale factor of the paths through the GlobalMinimax.
64  int XScaleFactor() const override;
65 
66  // Provides the (minimum) x scale factor to the network (of interest only to
67  // input units) so they can determine how to scale bounding boxes.
68  void CacheXScaleFactor(int factor) override;
69 
70  // Runs forward propagation of activations on the input line.
71  // See Network for a detailed discussion of the arguments.
72  void Forward(bool debug, const NetworkIO &input,
73  const TransposedArray *input_transpose, NetworkScratch *scratch,
74  NetworkIO *output) override;
75 
76  // Runs backward propagation of errors on the deltas line.
77  // See Network for a detailed discussion of the arguments.
78  bool Backward(bool debug, const NetworkIO &fwd_deltas,
79  NetworkScratch *scratch, NetworkIO *back_deltas) override;
80  // Creates and returns a Pix of appropriate size for the network from the
81  // image_data. If non-null, *image_scale returns the image scale factor used.
82  // Returns nullptr on error.
83  /* static */
84  static Image PrepareLSTMInputs(const ImageData &image_data,
85  const Network *network, int min_width,
86  TRand *randomizer, float *image_scale);
87  // Converts the given pix to a NetworkIO of height and depth appropriate to
88  // the given StaticShape:
89  // If depth == 3, convert to 24 bit color, otherwise normalized grey.
90  // Scale to target height, if the shape's height is > 1, or its depth if the
91  // height == 1. If height == 0 then no scaling.
92  // NOTE: It isn't safe for multiple threads to call this on the same pix.
93  static void PreparePixInput(const StaticShape &shape, const Image pix,
94  TRand *randomizer, NetworkIO *input);
95 
96 private:
97  void DebugWeights() override {
98  tprintf("Must override Network::DebugWeights for type %d\n", type_);
99  }
100 
101  // Input shape determines how images are dealt with.
102  StaticShape shape_;
103  // Cached total network x scale factor for scaling bounding boxes.
104  int cached_x_scale_;
105 };
106 
107 } // namespace tesseract.
108 
109 #endif // TESSERACT_LSTM_INPUT_H_
void tprintf(const char *format,...)
Definition: tprintf.cpp:41
StaticShape OutputShape([[maybe_unused]] const StaticShape &input_shape) const override
Definition: input.h:48
std::string spec() const override
Definition: input.h:35
StaticShape InputShape() const override
Definition: input.h:43
TESS_API Input(const std::string &name, int ni, int no)
Definition: input.cpp:30
void Forward(bool debug, const NetworkIO &input, const TransposedArray *input_transpose, NetworkScratch *scratch, NetworkIO *output) override
Definition: input.cpp:64
int XScaleFactor() const override
Definition: input.cpp:52
static Image PrepareLSTMInputs(const ImageData &image_data, const Network *network, int min_width, TRand *randomizer, float *image_scale)
Definition: input.cpp:81
void CacheXScaleFactor(int factor) override
Definition: input.cpp:58
bool Backward(bool debug, const NetworkIO &fwd_deltas, NetworkScratch *scratch, NetworkIO *back_deltas) override
Definition: input.cpp:71
~Input() override=default
static void PreparePixInput(const StaticShape &shape, const Image pix, TRand *randomizer, NetworkIO *input)
Definition: input.cpp:107
bool Serialize(TFile *fp) const override
Definition: input.cpp:40
bool DeSerialize(TFile *fp) override
Definition: input.cpp:45
NetworkType type_
Definition: network.h:300
const std::string & name() const
Definition: network.h:140
#define TESS_API
Definition: export.h:34