Mercurial > hgrepos > Python2 > PyMuPDF

comparison mupdf-source/thirdparty/tesseract/src/wordrec/params_model.cpp @ 2:b50eed0cc0ef upstream

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
ADD: MuPDF v1.26.7: the MuPDF source as downloaded by a default build of PyMuPDF 1.26.4. The directory name has changed: no version number in the expanded directory now.
author Franz Glasner <fzglas.hg@dom66.de>
date Mon, 15 Sep 2025 11:43:07 +0200
parents
children
comparison
equal deleted inserted replaced
1:1d09e1dec1d9 2:b50eed0cc0ef
1 ///////////////////////////////////////////////////////////////////////
2 // File: params_model.cpp
3 // Description: Trained language model parameters.
4 // Author: David Eger
5 //
6 // (C) Copyright 2012, 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.
16 //
17 ///////////////////////////////////////////////////////////////////////
18
19 #include "params_model.h"
20
21 #include <cctype>
22 #include <cmath>
23 #include <cstdio>
24
25 #include "bitvector.h"
26 #include "helpers.h" // for ClipToRange
27 #include "serialis.h" // for TFile
28 #include "tprintf.h"
29
30 namespace tesseract {
31
32 // Scale factor to apply to params model scores.
33 static const float kScoreScaleFactor = 100.0f;
34 // Minimum cost result to return.
35 static const float kMinFinalCost = 0.001f;
36 // Maximum cost result to return.
37 static const float kMaxFinalCost = 100.0f;
38
39 void ParamsModel::Print() {
40 for (int p = 0; p < PTRAIN_NUM_PASSES; ++p) {
41 tprintf("ParamsModel for pass %d lang %s\n", p, lang_.c_str());
42 for (unsigned i = 0; i < weights_vec_[p].size(); ++i) {
43 tprintf("%s = %g\n", kParamsTrainingFeatureTypeName[i], weights_vec_[p][i]);
44 }
45 }
46 }
47
48 void ParamsModel::Copy(const ParamsModel &other_model) {
49 for (int p = 0; p < PTRAIN_NUM_PASSES; ++p) {
50 weights_vec_[p] = other_model.weights_for_pass(static_cast<PassEnum>(p));
51 }
52 }
53
54 // Given a (modifiable) line, parse out a key / value pair.
55 // Return true on success.
56 bool ParamsModel::ParseLine(char *line, char **key, float *val) {
57 if (line[0] == '#') {
58 return false;
59 }
60 int end_of_key = 0;
61 while (line[end_of_key] && !(isascii(line[end_of_key]) && isspace(line[end_of_key]))) {
62 end_of_key++;
63 }
64 if (!line[end_of_key]) {
65 tprintf("ParamsModel::Incomplete line %s\n", line);
66 return false;
67 }
68 line[end_of_key++] = 0;
69 *key = line;
70 if (sscanf(line + end_of_key, " %f", val) != 1) {
71 return false;
72 }
73 return true;
74 }
75
76 // Applies params model weights to the given features.
77 // Assumes that features is an array of size PTRAIN_NUM_FEATURE_TYPES.
78 // The cost is set to a number that can be multiplied by the outline length,
79 // as with the old ratings scheme. This enables words of different length
80 // and combinations of words to be compared meaningfully.
81 float ParamsModel::ComputeCost(const float features[]) const {
82 float unnorm_score = 0.0;
83 for (int f = 0; f < PTRAIN_NUM_FEATURE_TYPES; ++f) {
84 unnorm_score += weights_vec_[pass_][f] * features[f];
85 }
86 return ClipToRange(-unnorm_score / kScoreScaleFactor, kMinFinalCost, kMaxFinalCost);
87 }
88
89 bool ParamsModel::Equivalent(const ParamsModel &that) const {
90 float epsilon = 0.0001f;
91 for (int p = 0; p < PTRAIN_NUM_PASSES; ++p) {
92 if (weights_vec_[p].size() != that.weights_vec_[p].size()) {
93 return false;
94 }
95 for (unsigned i = 0; i < weights_vec_[p].size(); i++) {
96 if (weights_vec_[p][i] != that.weights_vec_[p][i] &&
97 std::fabs(weights_vec_[p][i] - that.weights_vec_[p][i]) > epsilon) {
98 return false;
99 }
100 }
101 }
102 return true;
103 }
104
105 bool ParamsModel::LoadFromFp(const char *lang, TFile *fp) {
106 const int kMaxLineSize = 100;
107 char line[kMaxLineSize];
108 BitVector present;
109 present.Init(PTRAIN_NUM_FEATURE_TYPES);
110 lang_ = lang;
111 // Load weights for passes with adaption on.
112 std::vector<float> &weights = weights_vec_[pass_];
113 weights.clear();
114 weights.resize(PTRAIN_NUM_FEATURE_TYPES, 0.0f);
115
116 while (fp->FGets(line, kMaxLineSize) != nullptr) {
117 char *key = nullptr;
118 float value;
119 if (!ParseLine(line, &key, &value)) {
120 continue;
121 }
122 int idx = ParamsTrainingFeatureByName(key);
123 if (idx < 0) {
124 tprintf("ParamsModel::Unknown parameter %s\n", key);
125 continue;
126 }
127 if (!present[idx]) {
128 present.SetValue(idx, true);
129 }
130 weights[idx] = value;
131 }
132 bool complete = (present.NumSetBits() == PTRAIN_NUM_FEATURE_TYPES);
133 if (!complete) {
134 for (int i = 0; i < PTRAIN_NUM_FEATURE_TYPES; i++) {
135 if (!present[i]) {
136 tprintf("Missing field %s.\n", kParamsTrainingFeatureTypeName[i]);
137 }
138 }
139 lang_ = "";
140 weights.clear();
141 }
142 return complete;
143 }
144
145 bool ParamsModel::SaveToFile(const char *full_path) const {
146 const std::vector<float> &weights = weights_vec_[pass_];
147 if (weights.size() != PTRAIN_NUM_FEATURE_TYPES) {
148 tprintf("Refusing to save ParamsModel that has not been initialized.\n");
149 return false;
150 }
151 FILE *fp = fopen(full_path, "wb");
152 if (!fp) {
153 tprintf("Could not open %s for writing.\n", full_path);
154 return false;
155 }
156 bool all_good = true;
157 for (unsigned i = 0; i < weights.size(); i++) {
158 if (fprintf(fp, "%s %f\n", kParamsTrainingFeatureTypeName[i], weights[i]) < 0) {
159 all_good = false;
160 }
161 }
162 fclose(fp);
163 return all_good;
164 }
165
166 } // namespace tesseract