README.md

---
library_name: transformers
pipeline_tag: image-to-text
tags:
- chemistry
- image-to-text
- ocsr
- markush
- e-smiles2.0
datasets:
- UniParser/MolParser-7M
- UniParser/MolGallery
license: cc-by-nc-sa-4.0
---

# MolParser Mobile V2

<p align="center">
💻 <a href="https://github.com/dptech-corp/MolParser">GitHub</a> |
📘 <a href="https://github.com/dptech-corp/MolParser/blob/main/skills/molparser-extended-smiles/extended-smiles-spec.md">E-SMILES 2.0 Spec</a> |
📄 <a href="https://arxiv.org/abs/2609.05807">Report</a> |
🚀 <a href="https://huggingface.co/spaces/AI4Industry/molparser-mobile">Demo</a>
</p>

**MolParser-Mobile-V2** is a lightweight Optical Chemical Structure Recognition (OCSR) model that converts molecular structure images directly into **E-SMILES 2.0**. It upgrades MolParser-Mobile for broader recognition of structures found in chemical literature, especially complex Markush structures, while retaining a compact 10M parameter architecture.


## 🚀 What's New

* **E-SMILES 2.0 output** with substantially broader coverage of literature molecules and Markush structures.
* **Richer Markush type coverage** for literature molecules, including atom- and ring-indexed substituents, explicit dummy attachments, nested substructures, structural repeating units and polymers, virtual arcs, colored endpoint balls, and axial-chirality annotations.
* **384 × 384 input resolution**, increased from 224 × 224 in MolParser-Mobile.
* **384-token maximum output length**, increased from 256 tokens.
* **Improved recognition accuracy**, particularly for complex and stereochemical structures.

For notation details, examples, validation, normalization, substitution, and rendering utilities, see the [MolParser Repo](https://github.com/dptech-corp/MolParser) and the [E-SMILES specification](https://github.com/dptech-corp/MolParser/blob/main/skills/molparser-extended-smiles/extended-smiles-spec.md).


## 📊 Performance

Accuracy for MolParser-Mobile-V2 was measured with FP16 inference, greedy decoding, and batch size 512. Deltas are relative to MolParser-Mobile.

| Model                    | Parameters | Throughput (RTX 4090D) |     Uni-Parser Bench |             BioVista |          WildMol-10k |               USPTO |
| ------------------------ | ---------: | ---------------------: | -------------------: | -------------------: | -------------------: | ------------------: |
| MolParser-Mobile         |      9.98M |           1,520 Mol/s  |                0.823 |                0.801 |                0.734 |               0.836 |
| **MolParser-Mobile-V2**  |     10.00M |           1,296 Mol/s  | **0.850** (+0.027)   | **0.820** (+0.019)   | **0.762** (+0.028)   | **0.909** (+0.073)  |


## ⚡ Usage

### Option 1. MolParser Library (Recommended)

The [MolParser library](https://github.com/dptech-corp/MolParser) provides a convenient interface for molecule detection, recognition, E-SMILES 2.0 post-processing, and rendering.

Clone the repository and install the package:

```bash
git clone https://github.com/dptech-corp/MolParser.git
cd MolParser
pip install -e .
```

Then run:

```python
from molparser import MolParser

parser = MolParser(molparser_hf_repo="UniParser/MolParser-Mobile-V2", max_length=384) 

result = parser.parse("mol.png", rec_only=True)
```

To render the predicted E-SMILES as SVG or PNG, see [Render E-SMILES](https://github.com/dptech-corp/MolParser/tree/main#render-e-smiles):

```python
from pathlib import Path
from molparser import utils as mutils

raw = "*C(O)c1cc(C(=O)N(*)*)cc(-c2*ccc*2)c1<sep><a>0:CF3</a><a>9:R[3]</a><a>10:R[2]</a><a>14:X</a><a>18:Y</a><r>1:R[1]?1-3</r>"
svg_text = mutils.draw(raw, output_format="svg")
Path("molecule.svg").write_text(svg_text, encoding="utf-8")

png_bytes = mutils.draw(raw, output_format="png")
Path("molecule.png").write_bytes(png_bytes)
```

### Option 2. 🤗 Transformers

Load MolParser-Mobile-V2 directly with the Hugging Face transformers library.

```python
import torch
from PIL import Image
from transformers import AutoModelForImageTextToText, AutoProcessor

repo_id = "UniParser/MolParser-Mobile-V2"
device = "cuda" if torch.cuda.is_available() else "cpu"
dtype = torch.float16 if device == "cuda" else torch.float32

processor = AutoProcessor.from_pretrained(repo_id, trust_remote_code=True)
model = AutoModelForImageTextToText.from_pretrained(
    repo_id,
    dtype=dtype,
    trust_remote_code=True,
).to(device).eval()

image = Image.open("mol.png").convert("RGB")
inputs = processor(images=image, return_tensors="pt")
inputs = {k: v.to(device, dtype=dtype) for k, v in inputs.items()}

output_ids = model.generate(**inputs, max_length=384, num_beams=1, do_sample=False)
caption = processor.batch_decode(output_ids, skip_special_tokens=True)[0]
print(caption)
```


## 📜 License

### MolParser-Mobile-V2 Weight

The **MolParser-Mobile-V2 model weights** are provided for **non-commercial use only** under CC BY-NC-SA 4.0.

For commercial licensing, please contact **fangxi@dp.tech** or open a discussion on Hugging Face.

### MolParser Github Repo

The **MolParser** library (including E-SMILES post-processing and rendering) is available at https://github.com/dptech-corp/MolParser and is licensed under the **Apache License 2.0**, which permits commercial use, modification, and distribution, provided that the license and copyright notices are retained.

**Note:** Model weights, datasets, and third-party dependencies are subject to their respective licenses.

## 📖 Citation

If you use this model, please cite:

```
@article{fang2026molparserm,
  title={MolParser-Mobile: Ultrafast OCSR System for Large-Scale Chemical Literature Mining},
  author={Fang, Xi and Lu, Haocheng and Lyu, Han and Luo, Chengxiang and Zhang, Linfeng and Ke, Guolin},
  journal={arXiv preprint arXiv:2609.05807},
  year={2026}
}
```

```
@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}
}
```