Home - News ( United States | United Kingdom | Italy | Germany ) - Football scores

Showing content from https://tesseract-ocr.github.io/tessdoc/Planning.html below:

Tesseract release planning | tessdoc

Here we can plan the next releases of Tesseract.

Here are some ideas for future Tesseract releases.

• Modernize the code using C++11 (see discussions here and here).

• Use llvm’s tools: clang-format, clang-tidy, scan-build, sanitizers.

• Replace more Tesseract data types by C++ standard types (GenericVector, …), especially for the API.

• Add json (or xml) output format. It will be used for full ocr and for psm 2 - layout info only.

• Add option to use alternative binarization methods from leptonica.

• Add an option to output separate files for multipage input (out1.hocr, out2.hocr …).

• Add multi-threading option to the command line (openmp will be disabled at runtime in this mode).

• Explore the option to use Protocol Buffers or FlatBuffers for the traineddata.

• Improve error handling and don’t ignore return values from functions (see discussion).

• Replace tprintf etc. by advanced logging API with log levels.

Requirements (see also discussion):

Log levels:

• trace
• debug
• info
• warning
• error
• fatal

Related issues:

• https://github.com/tesseract-ocr/tesseract/issues/1338

Useful links:

• List of Open Source C++ logging libraries

See the release notes.

See also the discussion for issue #1423.

• Issues with the “bug” label (see list here)
• Noise characters recognized with bbox as the entire page #1192
• Segmentation fault when using integer models for LSTM training #1573
• Insufficient error message when output file cannot be created Issue 1424
• “no best words!!” on mixed language (fra+ara) items (see issue 235)
• mgr_.Init(traineddata_path.c_str()):Error:Assert failed: #1075 (see issue 1075)

• Script for installing only selected languages from github (see issue)

  https://github.com/zdenop/tessdata_downloader

Depending on available resources and opinions, these suggestions will either be added to the planning for the next or a future release or abandoned.

• Enhance –list-langs to show additional information for scripts and languages like legacy / LSTM, version
  

  This will make the command slower, because each file must be opened and parsed. Add this as –list-langs-details or as –list-lang-details for one language file based on lang-code?

• –list-langs should also display the directory it is using
• Fix the autotools build so that the debug mode uses -O0 as intended
• Add option to optionally select implementation for dot product (CPU, SSE, AVX, …)
• Relative includes for traineddata

  tessedit_load_sublangs should search for the sublangs relative to the parent, not starting in tessdata dir.

• More fixes for compiler warnings and issues reported by Coverity Scan
• Add a simple bash script for building tesseract
• New traineddata format

  In addition to the current proprietary format Tesseract could also support ZIP archives (see discussion).

  A possible implementation using libarchive is available, but needs more testing.

• “Training light” - Learning by doing (see issue)
• Modify text2image to use PrepareDistortedPix() #1052
• Schedule date

Tesseract 4.0 should be a full replacement for Tesseract 3.05 and have the same features when used with the old OCR engine (--oem 0). The following regressions still need verification (are they really regressions, or are they just missing features for LSTM):

• User Patterns (See issue)

Fixed in 4.1.0

These features still work with the old OCR engine (--oem 0), but are missing and desired for LSTM.

• #### Black list / White list (See issue). Here is a workaround.Fixed in 4.1.0.

Here we collect important issues and features for the release(s) following 4.0.0.

• Remove Legacy Tesseract Engine (see issue)
• ARM SIMD support for dot product #519
• Using OpenMP for dot product #983
• Remove deprecated code

  This does not include OpenCL or the old Tesseract engine.

• Tesseract creates output for missing input (see issue 1023).

  Mostly solved, but could be improved.

• Issue 1353: Patch for /training/tessopt.cpp (see pull request 13)

  It looks like it is not possible to run more than one training in the same process. The pull request describes a possible fix, but does not include a complete implementation (low priority).

RetroSearch is an open source project built by @garambo | Open a GitHub Issue

Search and Browse the WWW like it's 1997 | Search results from DuckDuckGo

HTML: 3.2 | Encoding: UTF-8 | Version: 0.7.3