---
task_categories:
- image-to-text
- text-to-image
language:
- en
pretty_name: flickr8k
size_categories:
- 1K<n<10K
---

You must download the dataset files manually. You can visit [this](https://github.com/jbrownlee/Datasets/releases/tag/Flickr8k) page or run `download.sh` to get files.

After, you can load dataset by referencing the directory:

```py
import datasets
ds = datasets.load_dataset("atasoglu/flickr8k-dataset", data_dir="data")
print(ds)
```

```
DatasetDict({
    train: Dataset({
        features: ['image_id', 'image_path', 'captions'],
        num_rows: 6000
    })
    test: Dataset({
        features: ['image_id', 'image_path', 'captions'],
        num_rows: 1000
    })
    validation: Dataset({
        features: ['image_id', 'image_path', 'captions'],
        num_rows: 1000
    })
})
```

I don't own the copyright of the images. Please [visit](https://forms.illinois.edu/sec/1713398) for more.