README.md

---
language:
- en
license: other
license_name: mixed-see-sources
task_categories:
- image-to-text
tags:
- chemistry
- cheminformatics
- ocsr
- optical-chemical-structure-recognition
- molecule-recognition
- smiles
- benchmark
pretty_name: OCSR Benchmarks
size_categories:
- 10K<n<100K
configs:
- config_name: ACS
  data_files:
  - split: test
    path: ACS/test-*
- config_name: CLEF
  data_files:
  - split: test
    path: CLEF/test-*
- config_name: ChemDraw
  data_files:
  - split: test
    path: ChemDraw/test-*
- config_name: Indigo
  data_files:
  - split: test
    path: Indigo/test-*
- config_name: JPO
  data_files:
  - split: test
    path: JPO/test-*
- config_name: Staker
  data_files:
  - split: test
    path: Staker/test-*
- config_name: UOB
  data_files:
  - split: test
    path: UOB/test-*
- config_name: USPTO
  data_files:
  - split: test
    path: USPTO/test-*
- config_name: USPTO-10K
  data_files:
  - split: test
    path: USPTO-10K/test-*
- config_name: WildMol-10K
  data_files:
  - split: test
    path: WildMol-10K/test-*
dataset_info:
- config_name: ACS
  features:
  - name: image_id
    dtype: string
  - name: image
    dtype: image
  - name: SMILES
    dtype: string
  splits:
  - name: test
    num_bytes: 5001836
    num_examples: 331
  download_size: 5006693
  dataset_size: 5001836
- config_name: CLEF
  features:
  - name: image_id
    dtype: string
  - name: image
    dtype: image
  - name: SMILES
    dtype: string
  splits:
  - name: test
    num_bytes: 3308074
    num_examples: 992
  download_size: 3295643
  dataset_size: 3308074
- config_name: ChemDraw
  features:
  - name: image_id
    dtype: string
  - name: image
    dtype: image
  - name: SMILES
    dtype: string
  splits:
  - name: test
    num_bytes: 28724502
    num_examples: 5719
  download_size: 28929307
  dataset_size: 28724502
- config_name: Indigo
  features:
  - name: image_id
    dtype: string
  - name: image
    dtype: image
  - name: SMILES
    dtype: string
  splits:
  - name: test
    num_bytes: 224451958
    num_examples: 5719
  download_size: 226875357
  dataset_size: 224451958
- config_name: JPO
  features:
  - name: image_id
    dtype: string
  - name: image
    dtype: image
  - name: SMILES
    dtype: string
  splits:
  - name: test
    num_bytes: 2862316
    num_examples: 449
  download_size: 2860290
  dataset_size: 2862316
- config_name: Staker
  features:
  - name: image_id
    dtype: string
  - name: image
    dtype: image
  - name: SMILES
    dtype: string
  splits:
  - name: test
    num_bytes: 143142981
    num_examples: 50000
  download_size: 143167707
  dataset_size: 143142981
- config_name: UOB
  features:
  - name: image_id
    dtype: string
  - name: image
    dtype: image
  - name: SMILES
    dtype: string
  splits:
  - name: test
    num_bytes: 19040066
    num_examples: 5740
  download_size: 19037801
  dataset_size: 19040066
- config_name: USPTO
  features:
  - name: image_id
    dtype: string
  - name: image
    dtype: image
  - name: SMILES
    dtype: string
  splits:
  - name: test
    num_bytes: 20247285
    num_examples: 5719
  download_size: 20281040
  dataset_size: 20247285
- config_name: USPTO-10K
  features:
  - name: image_id
    dtype: string
  - name: image
    dtype: image
  - name: SMILES
    dtype: string
  splits:
  - name: test
    num_bytes: 59935636
    num_examples: 9999
  download_size: 59096324
  dataset_size: 59935636
- config_name: WildMol-10K
  features:
  - name: image_id
    dtype: string
  - name: image
    dtype: image
  - name: SMILES
    dtype: string
  splits:
  - name: test
    num_bytes: 214403509
    num_examples: 9889
  download_size: 212929337
  dataset_size: 214403509
---

# OCSR Benchmarks

A collection of ten benchmark datasets for **Optical Chemical Structure Recognition (OCSR)** —
the task of converting chemical structure diagram images into machine-readable SMILES strings.

These benchmarks were used to evaluate the [COMO model](https://huggingface.co/Keylab/COMO)
(Closed-Loop Optical Molecule Recognition).

## Subsets

| Config | Split | Size | Domain |
|---|---|---|---|
| `CLEF` | test | 992 | Real |
| `JPO` | test | 449 | Real |
| `UOB` | test | 5,740 | Real |
| `USPTO` | test | 5719 | Real |
| `USPTO-10K` | test | 9,999 | Real |
| `Staker` | test | 50,000 | Real |
| `ACS` | test | 331 | Real |
| `WildMol-10K` | test | 9,889 | Real |
| `Indigo` | test | 5,719 | Synthetic |
| `ChemDraw` | test | 5,719 | Synthetic |

## Schema

Each sample has three fields:

| Field | Type | Description |
|---|---|---|
| `image_id` | `string` | Original identifier for the sample |
| `image` | `Image` | PNG image of the chemical structure diagram |
| `SMILES` | `string` | Ground-truth SMILES string |

## Usage

```python
from datasets import load_dataset

# Load a single benchmark
ds = load_dataset("Keylab/OCSR-Benchmarks", name="USPTO", split="test")
sample = ds[0]
sample["image"].show()   # PIL Image
print(sample["SMILES"])

# Iterate over all benchmarks
for config in ["CLEF", "JPO", "UOB", "USPTO", "USPTO-10K",
               "Staker", "ACS", "WildMol-10K", "Indigo", "ChemDraw"]:
    ds = load_dataset("Keylab/OCSR-Benchmarks", name=config, split="test")
    print(f"{config}: {len(ds)} samples")
```

## Bulk Download

Pre-packaged `.tar.gz` archives (images + CSV) are also available in the
[COMO model repository](https://huggingface.co/Keylab/COMO/tree/main/benchmarks)
for direct download without the `datasets` library.

## License

These benchmarks are collected from existing public OCSR datasets.
Please refer to the original sources for attribution and applicable terms:

| Dataset | Source |
|---------|--------|
| USPTO, CLEF, JPO, UOB, Staker | [Rajan et al., 2020](https://github.com/Kohulan/OCSR_Review), [Xiong et al., 2023](https://github.com/jiachengxiong/alpha-Extractor) |
| Indigo, ChemDraw, ACS, Staker | [Qian et al., 2023](https://github.com/thomas0809/MolScribe) |
| USPTO-10K | [Morin et al., 2023](https://huggingface.co/datasets/docling-project/USPTO-30K) |
| WildMol-10K | [Fang et al., 2025](https://github.com/orgs/Chem-Struct-ML/repositories) |


## Citation

If you use these benchmarks, please cite the COMO paper and the original benchmark sources:

```bibtex
@article{lyu2026closed,
  title={COMO: Closed-Loop Optical Molecule Recognition with Minimum Risk Training},
  author={Lyu, Zhuoqi and Ke, Qing},
  journal={arXiv preprint arXiv:2604.23546},
  year={2026}
}
```