https://github.com/duaibeom/chemOCR
DB-based Optical Chemical Structure Recognition
https://github.com/duaibeom/chemOCR
Last synced: about 24 hours ago
JSON representation
DB-based Optical Chemical Structure Recognition
- Host: GitHub
- URL: https://github.com/duaibeom/chemOCR
- Owner: duaibeom
- License: mit
- Created: 2022-08-30T12:29:26.000Z (over 2 years ago)
- Default Branch: main
- Last Pushed: 2022-09-12T08:46:48.000Z (over 2 years ago)
- Last Synced: 2024-11-08T02:32:39.825Z (5 months ago)
- Language: Python
- Size: 47.6 MB
- Stars: 7
- Watchers: 1
- Forks: 2
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
- awesome - duaibeom/chemOCR - DB-based Optical Chemical Structure Recognition (Python)
README
ChemOCR(OCSR)
DB(Differentiable Binarization)-based Optical Chemical Structure Recognition
---
```sh
streamlit run demo.py
```
### Backbone
- Swin
- ResNet
- MobileNetV3
## Parser
**Rule-based method**
## DATA
- ChEMBL: https://www.ebi.ac.uk/chembl/
### Training data sample
### Limitation
- Non-charged atom
- (Inner bridged) Complex ring is not possible.
- Specific resolution of image
- Small text atoms (like Iodin) are not recognized.
### TODO
- [x] Character recognition (pretrained model from MMOCR RCNN)
- [x] Web front-end (streamlit)
- [ ] Bond direction analysis
## References
1. https://github.com/MhLiao/DB
2. https://github.com/open-mmlab/mmocr
3. https://github.com/rdkit/rdkit