---
license: cc-by-nc-4.0
task_categories:
- image-to-text
- visual-question-answering
language:
- en
tags:
- multimodal
- vision
- benchmark
size_categories:
- 1K<n<10K
---

# MultiRef Dataset

This repository contains two datasets for multi-image reference tasks:

## MultiRef-Bench-Synthetic (900 samples)
- **images/**: Processed images for the benchmark
- **original_images/**: Original unprocessed images
- **benchmark990v3.json**: Benchmark data with 990 entries (first 900 used)

## Multi-Image-Benchmark (1000 samples)
- **compressed_images/**: Compressed images for the benchmark
- **final_1000_prompts_taxonomy.json**: Taxonomy data with 1000 prompts

## Usage

```python
from datasets import load_dataset

# Load the dataset
dataset = load_dataset("wsnHowest/MultiRef-dataset")
```

## File Structure
```
MultiRef-dataset/
├── MultiRef-Bench-Synthetic/
│   ├── images/
│   ├── original_images/
│   └── benchmark990v3.json
└── multi_image_benchmark/
    ├── compressed_images/
    └── final_1000_prompts_taxonomy.json
```

## License
cc-by-nc-4.0