This fixes warnings from the Intel compiler:
src/textord/cjkpitch.cpp(319): warning #177:
function "<unnamed>::FPRow::good_gaps" was declared but never referenced
src/textord/cjkpitch.cpp(383): warning #177:
function "<unnamed>::FPRow::is_bad" was declared but never referenced
src/textord/cjkpitch.cpp(387): warning #177:
function "<unnamed>::FPRow::is_unknown" was declared but never referenced
Signed-off-by: Stefan Weil <sw@weilnetz.de>
This fixes a warning from the Intel compiler:
src/textord/cjkpitch.cpp(79): warning #177:
function "<unnamed>::SimpleStats::maximum" was declared
but never referenced
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Instrumented code throws this runtime error during OCR:
../../src/api/baseapi.cpp:1616:5: runtime error: load of value 128,
which is not a valid value for type 'bool'
../../src/api/baseapi.cpp:1627:5: runtime error: load of value 128,
which is not a valid value for type 'bool'
If there is no font information (typical for Tesseract with a LSTM model),
the font attributes got random values resulting in wrong hOCR output.
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Instrumented code throws this runtime error during OCR:
../../src/ccstruct/matrix.h:84:11: runtime error:
null pointer passed as argument 2, which is declared to never be null
Signed-off-by: Stefan Weil <sw@weilnetz.de>
All also a C++ implementation with more aggressive compiler options
which is optimized for the CPU where the software was built.
It is now possible to select the function used for the dot product
with -c dotproduct=FUNCTION where FUNCTION can be one of those values:
* auto selection based on detected hardware (default)
* generic C++ code with default compiler options
* native C++ code optimized for build host
* avx optimized code for AVX
* sse optimized code for SSE
Signed-off-by: Stefan Weil <sw@weilnetz.de>
This reduces the code size for intsimdmatrixavx2 from 2700 to 2668
and slightly improves the performance for fast models with AVX2.
Signed-off-by: Stefan Weil <sw@weilnetz.de>
This improves performace for the "best" models because it
avoids function calls.
The compiler also knows the passed values for the parameters
add_bias_fwd and skip_bias_back.
Signed-off-by: Stefan Weil <sw@weilnetz.de>
This is a lightweight, semi-Pythonic conversion of tesstrain.sh that currently
supports only LSTM and not the Tesseract 3 training mode.
I attempted to keep source changes minimal so it would be easy to compare
bash to Python in code review and confirm equivalence.
Python 3.6+ is required. Ubuntu 18.04 ships Python 3.6 and it is a mandatory
package (the package manager is also written in Python), so it is available
in the baseline Tesseract 4.0 system.
There are minor output and behavioral changes, and advantages. Python's loggingis used. Temporary files are only deleted on success, so they can be inspected
if training files. Console output is more terse and the log file is more
verbose. And there are progress bars! (The python3-tqdm package is required.)
Where tesstrain.sh would sometimes fail without explanation and return an error
code of 1, it is much easier to find the point of failure in this version.
That was also the main motivation for this work.
Argument checking is also more comprehensive.
The local variable k should be 10 ^ (number of digits after comma),
but will overflow when there are more than 9 digits after the comma
because an int value cannot store 10000000000.
This results in wrong double values read from .tr files for example
(or in a runtime exception if Tesseract was compiled with -ftrapv).
Using uint64_t does not fix the general problem but allows more digits
which should be sufficient for the data read by Tesseract.
Signed-off-by: Stefan Weil <sw@weilnetz.de>
shellcheck warning:
In /tesseract/src/training/tesstrain_utils.sh line 209:
TIMESTAMP=`date +%Y-%m-%d`
^-- SC2006: Use $(..) instead of legacy `..`.
Signed-off-by: Stefan Weil <sw@weilnetz.de>
The commit 10f2c45c00 unified the usage of mktemp, but with a
incorrect bash syntax and unnecessary definition of LANG_CODE
and TIMESTAMP. This patch fixes the above problems.
