---
annotations_creators:
- expert-generated
- machine-generated
language:
- en
license:
- other
multilinguality:
- monolingual
size_categories:
- 1K<n<10K
task_categories:
- image-to-text
- visual-question-answering
task_ids:
- visual-question-answering
---

# RobustRDP-RawTrainData

Raw training data for the paper: *RobustRDP: Advancing Reaction Diagram Parsing via Synthetic-to-Real Data Scaling and Robustness-Oriented Training*.

## Dataset Structure

```
raw_train_data/
├── images_train.zip          # Zipped training images (4,240 images)
└── labels_train.zip          # Zipped training labels (4,240 JSON files)
```

## Data Composition

The dataset contains a total of **4,240** reaction diagram images and their corresponding annotations:

| Source | Images | Description |
|--------|--------|-------------|
| **RxnLabel** | 3,000 | Manually annotated reaction diagrams using the RxnLabel annotation platform |
| **RxnScribe** | 1,240 | Reaction diagrams from the RxnScribe dataset |

## Data Format

### Images (`images_train/`)
PNG images of reaction diagrams extracted from chemistry literature.

### Labels (`labels_train/`)
Each image has a corresponding JSON label file containing the parsed reaction annotations with special tokens:
- `<rxn>`: Start of a reaction
- `<rct>`: Reactants section
- `<cnd>`: Conditions section
- `<prd>`: Products section
- `<mol>`: Molecule entity
- `<txt>`: Text entity

## Data Generation

- **RxnLabel (3,000 images)**: Manually annotated using [RxnLabel](https://github.com/jaydetang/RxnLabel), a web-based reaction diagram annotation platform.
- **RxnScribe (1,240 images)**: Sourced from the [RxnScribe](https://github.com/thomas0809/RxnScribe) dataset of reaction diagrams from chemistry publications.

## Related Resources

- **Model**: [RobustRDP](https://huggingface.co/Jingcz/RobustRDP)
- **Processed Data**: [RobustRDP-ProcessedTrainData](https://huggingface.co/datasets/Jingcz/RobustRDP-ProcessedTrainData)
- **Code**: [RobustRDP GitHub](https://github.com/jaydetang/RobustRDP)
- **Annotation Platform**: [RxnLabel](https://github.com/jaydetang/RxnLabel)