forked from tesseract-ocr/tesseract
-
Notifications
You must be signed in to change notification settings - Fork 1
/
lang_model_test.cc
201 lines (191 loc) · 9.68 KB
/
lang_model_test.cc
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
// (C) Copyright 2017, Google Inc.
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
// http://www.apache.org/licenses/LICENSE-2.0
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include <string> // for std::string
#include "gmock/gmock.h" // for testing::ElementsAreArray
#include "include_gunit.h"
#include "lang_model_helpers.h"
#include "log.h" // for LOG
#include "lstmtrainer.h"
#include "unicharset_training_utils.h"
namespace tesseract {
std::string TestDataNameToPath(const std::string &name) {
return file::JoinPath(TESTING_DIR, name);
}
// This is an integration test that verifies that CombineLangModel works to
// the extent that an LSTMTrainer can be initialized with the result, and it
// can encode strings. More importantly, the test verifies that adding an extra
// character to the unicharset does not change the encoding of strings.
TEST(LangModelTest, AddACharacter) {
constexpr char kTestString[] = "Simple ASCII string to encode !@#$%&";
constexpr char kTestStringRupees[] = "ASCII string with Rupee symbol ₹";
// Setup the arguments.
std::string script_dir = LANGDATA_DIR;
std::string eng_dir = file::JoinPath(script_dir, "eng");
std::string unicharset_path = TestDataNameToPath("eng_beam.unicharset");
UNICHARSET unicharset;
EXPECT_TRUE(unicharset.load_from_file(unicharset_path.c_str()));
std::string version_str = "TestVersion";
file::MakeTmpdir();
std::string output_dir = FLAGS_test_tmpdir;
LOG(INFO) << "Output dir=" << output_dir << "\n";
std::string lang1 = "eng";
bool pass_through_recoder = false;
// If these reads fail, we get a warning message and an empty list of words.
std::vector<std::string> words = split(ReadFile(file::JoinPath(eng_dir, "eng.wordlist")), '\n');
EXPECT_GT(words.size(), 0);
std::vector<std::string> puncs = split(ReadFile(file::JoinPath(eng_dir, "eng.punc")), '\n');
EXPECT_GT(puncs.size(), 0);
std::vector<std::string> numbers = split(ReadFile(file::JoinPath(eng_dir, "eng.numbers")), '\n');
EXPECT_GT(numbers.size(), 0);
bool lang_is_rtl = false;
// Generate the traineddata file.
EXPECT_EQ(0, CombineLangModel(unicharset, script_dir, version_str, output_dir, lang1,
pass_through_recoder, words, puncs, numbers, lang_is_rtl, nullptr,
nullptr));
// Init a trainer with it, and encode kTestString.
std::string traineddata1 = file::JoinPath(output_dir, lang1, lang1) + ".traineddata";
LSTMTrainer trainer1;
trainer1.InitCharSet(traineddata1);
std::vector<int> labels1;
EXPECT_TRUE(trainer1.EncodeString(kTestString, &labels1));
std::string test1_decoded = trainer1.DecodeLabels(labels1);
std::string test1_str(&test1_decoded[0], test1_decoded.length());
LOG(INFO) << "Labels1=" << test1_str << "\n";
// Add a new character to the unicharset and try again.
int size_before = unicharset.size();
unicharset.unichar_insert("₹");
SetupBasicProperties(/*report_errors*/ true, /*decompose (NFD)*/ false, &unicharset);
EXPECT_EQ(size_before + 1, unicharset.size());
// Generate the traineddata file.
std::string lang2 = "extended";
EXPECT_EQ(EXIT_SUCCESS, CombineLangModel(unicharset, script_dir, version_str, output_dir, lang2,
pass_through_recoder, words, puncs, numbers, lang_is_rtl,
nullptr, nullptr));
// Init a trainer with it, and encode kTestString.
std::string traineddata2 = file::JoinPath(output_dir, lang2, lang2) + ".traineddata";
LSTMTrainer trainer2;
trainer2.InitCharSet(traineddata2);
std::vector<int> labels2;
EXPECT_TRUE(trainer2.EncodeString(kTestString, &labels2));
std::string test2_decoded = trainer2.DecodeLabels(labels2);
std::string test2_str(&test2_decoded[0], test2_decoded.length());
LOG(INFO) << "Labels2=" << test2_str << "\n";
// encode kTestStringRupees.
std::vector<int> labels3;
EXPECT_TRUE(trainer2.EncodeString(kTestStringRupees, &labels3));
std::string test3_decoded = trainer2.DecodeLabels(labels3);
std::string test3_str(&test3_decoded[0], test3_decoded.length());
LOG(INFO) << "labels3=" << test3_str << "\n";
// Copy labels1 to a std::vector, renumbering the null char to match trainer2.
// Since Tensor Flow's CTC implementation insists on having the null be the
// last label, and we want to be compatible, null has to be renumbered when
// we add a class.
int null1 = trainer1.null_char();
int null2 = trainer2.null_char();
EXPECT_EQ(null1 + 1, null2);
std::vector<int> labels1_v(labels1.size());
for (unsigned i = 0; i < labels1.size(); ++i) {
if (labels1[i] == null1) {
labels1_v[i] = null2;
} else {
labels1_v[i] = labels1[i];
}
}
EXPECT_THAT(labels1_v, testing::ElementsAreArray(&labels2[0], labels2.size()));
// To make sure we we are not cheating somehow, we can now encode the Rupee
// symbol, which we could not do before.
EXPECT_FALSE(trainer1.EncodeString(kTestStringRupees, &labels1));
EXPECT_TRUE(trainer2.EncodeString(kTestStringRupees, &labels2));
}
// Same as above test, for hin instead of eng
TEST(LangModelTest, AddACharacterHindi) {
constexpr char kTestString[] = "हिन्दी में एक लाइन लिखें";
constexpr char kTestStringRupees[] = "हिंदी में रूपये का चिन्ह प्रयोग करें ₹१००.००";
// Setup the arguments.
std::string script_dir = LANGDATA_DIR;
std::string hin_dir = file::JoinPath(script_dir, "hin");
std::string unicharset_path = TestDataNameToPath("hin_beam.unicharset");
UNICHARSET unicharset;
EXPECT_TRUE(unicharset.load_from_file(unicharset_path.c_str()));
std::string version_str = "TestVersion";
file::MakeTmpdir();
std::string output_dir = FLAGS_test_tmpdir;
LOG(INFO) << "Output dir=" << output_dir << "\n";
std::string lang1 = "hin";
bool pass_through_recoder = false;
// If these reads fail, we get a warning message and an empty list of words.
std::vector<std::string> words = split(ReadFile(file::JoinPath(hin_dir, "hin.wordlist")), '\n');
EXPECT_GT(words.size(), 0);
std::vector<std::string> puncs = split(ReadFile(file::JoinPath(hin_dir, "hin.punc")), '\n');
EXPECT_GT(puncs.size(), 0);
std::vector<std::string> numbers = split(ReadFile(file::JoinPath(hin_dir, "hin.numbers")), '\n');
EXPECT_GT(numbers.size(), 0);
bool lang_is_rtl = false;
// Generate the traineddata file.
EXPECT_EQ(0, CombineLangModel(unicharset, script_dir, version_str, output_dir, lang1,
pass_through_recoder, words, puncs, numbers, lang_is_rtl, nullptr,
nullptr));
// Init a trainer with it, and encode kTestString.
std::string traineddata1 = file::JoinPath(output_dir, lang1, lang1) + ".traineddata";
LSTMTrainer trainer1;
trainer1.InitCharSet(traineddata1);
std::vector<int> labels1;
EXPECT_TRUE(trainer1.EncodeString(kTestString, &labels1));
std::string test1_decoded = trainer1.DecodeLabels(labels1);
std::string test1_str(&test1_decoded[0], test1_decoded.length());
LOG(INFO) << "Labels1=" << test1_str << "\n";
// Add a new character to the unicharset and try again.
int size_before = unicharset.size();
unicharset.unichar_insert("₹");
SetupBasicProperties(/*report_errors*/ true, /*decompose (NFD)*/ false, &unicharset);
EXPECT_EQ(size_before + 1, unicharset.size());
// Generate the traineddata file.
std::string lang2 = "extendedhin";
EXPECT_EQ(EXIT_SUCCESS, CombineLangModel(unicharset, script_dir, version_str, output_dir, lang2,
pass_through_recoder, words, puncs, numbers, lang_is_rtl,
nullptr, nullptr));
// Init a trainer with it, and encode kTestString.
std::string traineddata2 = file::JoinPath(output_dir, lang2, lang2) + ".traineddata";
LSTMTrainer trainer2;
trainer2.InitCharSet(traineddata2);
std::vector<int> labels2;
EXPECT_TRUE(trainer2.EncodeString(kTestString, &labels2));
std::string test2_decoded = trainer2.DecodeLabels(labels2);
std::string test2_str(&test2_decoded[0], test2_decoded.length());
LOG(INFO) << "Labels2=" << test2_str << "\n";
// encode kTestStringRupees.
std::vector<int> labels3;
EXPECT_TRUE(trainer2.EncodeString(kTestStringRupees, &labels3));
std::string test3_decoded = trainer2.DecodeLabels(labels3);
std::string test3_str(&test3_decoded[0], test3_decoded.length());
LOG(INFO) << "labels3=" << test3_str << "\n";
// Copy labels1 to a std::vector, renumbering the null char to match trainer2.
// Since Tensor Flow's CTC implementation insists on having the null be the
// last label, and we want to be compatible, null has to be renumbered when
// we add a class.
int null1 = trainer1.null_char();
int null2 = trainer2.null_char();
EXPECT_EQ(null1 + 1, null2);
std::vector<int> labels1_v(labels1.size());
for (unsigned i = 0; i < labels1.size(); ++i) {
if (labels1[i] == null1) {
labels1_v[i] = null2;
} else {
labels1_v[i] = labels1[i];
}
}
EXPECT_THAT(labels1_v, testing::ElementsAreArray(&labels2[0], labels2.size()));
// To make sure we we are not cheating somehow, we can now encode the Rupee
// symbol, which we could not do before.
EXPECT_FALSE(trainer1.EncodeString(kTestStringRupees, &labels1));
EXPECT_TRUE(trainer2.EncodeString(kTestStringRupees, &labels2));
}
} // namespace tesseract