| --- |
| dataset_info: |
| features: |
| - name: image |
| dtype: image |
| - name: E-SMILES |
| dtype: string |
| splits: |
| - name: train |
| num_bytes: 195826797222 |
| num_examples: 10327534 |
| download_size: 192734235604 |
| dataset_size: 195826797222 |
| configs: |
| - config_name: default |
| data_files: |
| - split: train |
| path: data/train-* |
| --- |
| |
| # MolGallery: A 10 Million-Scale Real-World Dataset for Optical Chemical Structure Recognition |
|
|
| ## Dataset Summary |
|
|
| **MolGallery** is a large-scale dataset containing **10 million** chemical structure images sourced from real-world documents, including patents, scientific literature, books and website. |
|
|
| Unlike synthetic datasets, MolGallery captures the complex noise, diverse rendering styles, and scanning artifacts found in genuine chemical documents. The labels are provided in **E-SMILES** format, following the notation standards defined in the [MolParser](https://arxiv.org/abs/2411.11098) paper. |
|
|
|
|
| ## 🏷️ Pseudo-Labeling Pipeline |
|
|
| We use inference consistency across **two different weights of MolParser 1.0** together with predictions from the **MolScribe** model as a confidence criterion. Molecules with consistent predictions across these models are selected as **high-confidence samples** and used as **pseudo-labels**. |
|
|
| > ⚠️ **Notice** |
| > |
| > Pseudo-labels may contain a small proportion of errors. Therefore, they are primarily recommended for **pretraining**. |
| > For later training stages or the final training iterations, it is advisable to use **higher-quality human annotation data** to ensure optimal model performance. |
|
|
| ## 📜 License |
|
|
| This dataset is provided for **non-commercial use only**. |
|
|
| For commercial use, please contact: [fangxi@dp.tech](mailto:fangxi@dp.tech) or add a discussion in HuggingFace. |
|
|
|
|
| ## 📖 Citation |
|
|
| ``` |
| MolGallery Dataset Report Comming Soon! |
| ``` |
|
|
| If you use this datasets in your work, please cite: |
|
|
| ``` |
| @article{fang2025uniparser, |
| title={Uni-Parser Technical Report}, |
| author={Fang, Xi and Tao, Haoyi and Yang, Shuwen and Huang, Chaozheng and Zhong, Suyang and Lu, Haocheng and Lyu, Han and Li, Xinyu and Zhang, Linfeng and Ke, Guolin}, |
| journal={arXiv preprint arXiv:2512.15098}, |
| year={2025} |
| } |
| ``` |
|
|
| ``` |
| @inproceedings{fang2025molparser, |
| title={Molparser: End-to-end visual recognition of molecule structures in the wild}, |
| author={Fang, Xi and Wang, Jiankun and Cai, Xiaochen and Chen, Shangqian and Yang, Shuwen and Tao, Haoyi and Wang, Nan and Yao, Lin and Zhang, Linfeng and Ke, Guolin}, |
| booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision}, |
| pages={24528--24538}, |
| year={2025} |
| } |
| ``` |
|
|
