COCR

COCR is designed to convert an image of handwriting chemical structure to graph of that molecule.

COCR, Optical Character Recognition for Chemical Structures, was once a demo for my undergraduate graduation thesis in 2021.6. It brings OCSR(optical chemical structure recognition) capability into handwriting cases.

Symbol	String	Ring	Solid-	Hash-	Wavy-	Single-	Double-	Triple-
Appearance	(CH2)2COOEt	⏣	▲	△	~~	/	//	///
Status	✔️	✔️	✔️	✔️	✔️	✔️	✔️	✔️

COCR is developed under Qt framework. It handles images with YOLO and CRNN models using opencv or ncnn backend.

Online Demo

• try on chromium-based web browser like Chrome and Edge: https://xuguodong1999.github.io/COCR.html

Supported platforms:

• Windows
• Android
• WebAssembly
• Mac OS
• Linux

News

• [2021/06] COCR v1.1 released with support for strings and wavy-bonds.

v1.1.on.Ubuntu.mp4

• [2021/02] COCR v1.0 released for simple cases.

Input	Detection	Render

1. Support single element symbols: C、H、O、N、P、B、S、F、Cl、Br、I.
2. Support bond types: single, double, triple, hash wedge, solid wedge, circle.

Build

System Requirements

1. CMake (At least 3.22), Ninja
2. Git (For submodule clone)
3. Qt 6 (Test with 6.5.2)
4. Latest C++ Compiler (Test with GCC-11.x, MSVC-v17.x, Clang-16.x)

Qt Beginner's Guide

• for windows developer: have 7zip, wget and git-bash ready, or simply use WSL:

# download mingw, msvc binary
wget -r -np -nH -e robots=off http://mirrors.nju.edu.cn/qt/online/qtsdkrepository/windows_x86/desktop/qt6_652/
# extract all 7z files to a folder called "result"
find . -name '*.7z' -exec 7z x {} -aos -o./result \;

• for linux developer:

# download gcc_64 binary
wget -r -np -nH -e robots=off http://mirrors.nju.edu.cn/qt/online/qtsdkrepository/linux_x64/desktop/qt6_652/
# extract all 7z files to a folder called "result"
find . -name '*.7z' -exec 7z x {} -aos -o./result \;

• for macos developer:

# download clang_64 binary
wget -r -np -nH -e robots=off http://mirrors.nju.edu.cn/qt/online/qtsdkrepository/mac_x64/desktop/qt6_652/
# extract all 7z files to a folder called "result"
find . -name '*.7z' -exec 7z x {} -aos -o./result \;

after downloading and extracting, the folder structure looks like:

result/6.5.2
└── gcc_64
OR
result/6.5.2
└── msvc2019_64
OR
result/6.5.2
└── clang_64

Then copy gcc_64 or msvc2019_64 or clang_64 to your ideal library path and it's OK.

---

Get the Code

git clone https://github.com/xuguodong1999/COCR.git --branch main --single-branch --recursive

or using ssh

git clone git@github.com:xuguodong1999/COCR.git --branch main --single-branch --recursive

All third-party libraries except Qt are in third_party directory, including boost, openbabel, rdkit, opencv, etc.

Only a minimal set of source codes is kept, and you can find custom changes in patch files under third_party directory.

Nearly all build scripts for third-party libraries have been rewritten, to make cross-build, bugfix and feature hack easier.

Compile

Building COCR project is the same as common Qt 6 projects, the basic step is:

mkdir build && cd build
cmake .. -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON -DCMAKE_PREFIX_PATH=path/to/Qt6/binary/dir
cmake --build . --parallel --target COCR

for example, on linux desktop,

cmake .. -G Ninja -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON -DCMAKE_PREFIX_PATH=~/shared/Qt/6.5.2/gcc_64

More custom flags to control SIMD, Release Info, etc.

FLAG	VALUE	USAGE
XGD_USE_OPENMP	ON/OFF	enable openmp, default auto-detect
XGD_USE_VK	ON/OFF	enable vulkan, default auto-detect
XGD_USE_CCACHE	ON/OFF	enable ccache, default auto-detect
XGD_FLAG_MARCH_NATIVE	ON/OFF	add -march=native for gcc compiler, default ON
XGD_FLAG_WASM_SIMD128	ON/OFF	add -msimd128 for wasm, default ON
XGD_BUILD_WITH_GRADLE	ON/OFF	disable custom cmake directory layout for android studio cmake plugin, default OFF
XGD_OPT_RC	ON/OFF	add release info for products, default OFF
XGD_NO_DEBUG_CONSOLE	ON/OFF	hide debug console, default OFF
XGD_OPT_ARCH_X86	ON/O