---
dataset_info:
  features:
  - name: file_name
    dtype: string
  - name: image
    dtype: image
  - name: refs
    sequence: string
  - name: mt
    dtype: string
  - name: human_score
    dtype: float64
  splits:
  - name: train
    num_bytes: 4783117925.25
    num_examples: 26382
  - name: valid
    num_bytes: 593047697.75
    num_examples: 3298
  - name: test
    num_bytes: 595467629.75
    num_examples: 3298
  download_size: 5956104723
  dataset_size: 5971633252.75
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: valid
    path: data/valid-*
  - split: test
    path: data/test-*
---
# Nebula Dataset

[ACCV 2024] DENEB: A Hallucination-Robust Automatic Evaluation Metric for Image Captioning  
[![arXiv](https://img.shields.io/badge/arXiv-2409.19255-B31B1B)](https://arxiv.org/abs/2409.19255)
[![GitHub](https://img.shields.io/badge/GitHub-DENEB-181717?logo=github)](https://github.com/Ka2ukiMatsuda/DENEB)


## Usage

```python
>>> from datasets import load_dataset
>>> nebula = load_dataset("Ka2ukiMatsuda/Nebula")
>>> print(nebula)
DatasetDict({
    train: Dataset({
        features: ['file_name', 'image', 'refs', 'mt', 'human_score'],
        num_rows: 26382
    })
    valid: Dataset({
        features: ['file_name', 'image', 'refs', 'mt', 'human_score'],
        num_rows: 3298
    })
    test: Dataset({
        features: ['file_name', 'image', 'refs', 'mt', 'human_score'],
        num_rows: 3298
    })
})
```

## Citation

```bash
@inproceedings{matsuda2024deneb,
  title={DENEB: A Hallucination-Robust Automatic Evaluation Metric for Image Captioning},
  author={Kazuki Matsuda and Yuiga Wada and Komei Sugiura},
  booktitle={Proceedings of the Asian Conference on Computer Vision (ACCV)},
  year={2024},
  pages={3570--3586}
}
```