This is the official repository for the paper OCMR Is All You Need for Chemical Structure Recognition

Usage

1. Pull image from dockerhub. This image contains the tools of OCMR/MolVec/OSRA/Imago.

**docker pull limingisafish/ocmr:v4**

2. Starting the docker container on local port 1234.

docker run --rm -ti -p 1234:5632 limingisafish/ocmr:v4

3. Translate molecular image to SMILES.

python test.py

## {'imago': 'N#Cc1ccccc1C#N', 'molvec': 'N#CC1=C(C#N)C=CC=C1', 'osra': 'N#Cc1ccccc1C#N', 'ocmr': 'N#CC1=C(C#N)C=CC=C1'}

Benchmark Test

Environment

conda env create -f requirement.yml
conda activate ocmr_bench

Test PATENT dataset

1. Path to PATENT dataset

|----Data/
      |----Patent/
|----Evaluator_patent.py

2. run testing

python Evaluator_patent.py

3 the result will be saved in Patent_res.csv.

OCMR acc：415/520
MolVec acc：318/520
OSRA acc：342/520
Imago acc：120/520

Test UOB，CLEF，USPTO，JPO benchmark

|----Data/
      |----public_data/
            |----UOB/
            |----MolVec/
            |----OSRA/
            |----Imago/
|----Evaluator_public.py

2. run testing

python Evaluator_benchmarks.py

1. the result will be saved in all_result_public.xlsx and public_res.csv.

	SMI	Tanimoto
Group
CLEF	0.650202	0.931824
JPO	0.604444	0.879941
UOB	0.859930	0.980701
USPTO	0.746284	0.950304