intro/flickr8k

---
language:
- en
license: cc0-1.0
task_categories:
- image-to-text
- text-generation
pretty_name: Flickr8k Captions With Splits
size_categories:
- 1K<n<10K
configs:
- config_name: default
data_files:
- split: train
path: train/**
- split: dev
path: dev/**
- split: test
path: test/**
dataset_info:
features:
- name: image
dtype: image
- name: file_name
dtype: string
- name: split
dtype: string
- name: caption_0
dtype: string
- name: caption_1
dtype: string
- name: caption_2
dtype: string
- name: caption_3
dtype: string
- name: caption_4
dtype: string
---

# Flickr8k Captions With Splits

This dataset repackages the local Flickr8k image-caption corpus into a Hugging Face `imagefolder` layout with per-split `metadata.csv` files.

## Structure

- `train/`, `dev/`, and `test/` each contain image files and a `metadata.csv`.
- Each row represents one image with five captions.
- The `image` feature is inferred from `file_name` in each `metadata.csv`.

## Columns

- `image`: image file loaded by the Hugging Face Datasets library
- `file_name`: image filename stored as a string and used by `imagefolder` to map rows to image files
- `split`: one of `train`, `dev`, `test`
- `caption_0` to `caption_4`: the five raw Flickr8k captions for the image

## License

- CC0: Public Domain

## Counts

- Total included images: 8000
- Train images: 6000
- Dev images: 1000
- Test images: 1000
- Excluded caption entries without split or image file: 92

## Source Mapping

- Split assignments come from `Flickr_8k.trainImages.txt`, `Flickr_8k.devImages.txt`, and `Flickr_8k.testImages.txt`.
- Captions come from `Flickr8k.token.txt`, using `#0` through `#4` as `caption_0` through `caption_4`.

## Notes

- The original caption file contains 8,092 image-caption groups.
- Only the 8,000 images with explicit split assignments are included in this Hugging Face-ready dataset.
- Excluded rows are recorded in `excluded_rows.csv`.
- Missing local image files are recorded in `missing_image_rows.csv`.

## Loading Example

```python
from datasets import load_dataset

dataset = load_dataset("intro/flickr8k")
print(dataset["train"][0])
```

## Citation

If you use this dataset, please cite:

```bibtex
@article{hodosh2013framing,
title={Framing Image Description as a Ranking Task: Data, Models and Evaluation Metrics},
author={Hodosh, Micah and Young, Peter and Hockenmaier, Julia},
journal={Journal of Artificial Intelligence Research},
volume={47},
pages={853--899},
year={2013},
url={http://www.jair.org/papers/paper3994.html}
}
```