Datasets:

HiTZ
/

laion-eus

language:
  - eu
license: cc-by-4.0
size_categories:
  - 100K<n<1M
task_categories:
  - image-to-text
pretty_name: LAION-eus
dataset_info:
  features:
    - name: url
      dtype: string
    - name: similarity
      dtype: float64
    - name: hash
      dtype: int64
    - name: pwatermark
      dtype: float32
    - name: punsafe
      dtype: float32
    - name: lang_laion
      dtype: string
    - name: caption
      dtype: string
    - name: key
      dtype: string
    - name: status
      dtype: string
    - name: error_message
      dtype: string
    - name: width
      dtype: int32
    - name: height
      dtype: int32
    - name: original_width
      dtype: int32
    - name: original_height
      dtype: int32
    - name: exif
      dtype: string
    - name: md5
      dtype: string
    - name: lang
      dtype: string
    - name: lang_prob
      dtype: float64
  splits:
    - name: train
      num_bytes: 89834102
      num_examples: 221376
  download_size: 50506034
  dataset_size: 89834102
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

This dataset contains a subset of laion/relaion2B-multi-research where captions are written in Basque. Although language is already detected for its captions, we used another language detector Mike0307/multilingual-e5-language-detection and a high certainty threshold (p > 0.98) to better filter the instances in Basque.

Apart from the same attributes found in laion/relaion2B-multi-research, we add the following to each instance.

"lang_laion": language originally detected by the classifier used in laion/relaion2B-multi-research.
"lang": language detected by Mike0307/multilingual-e5-language-detection. This is always 'Basque'.
"lang_prob": probability of the caption to be written in Basque given by Mike0307/multilingual-e5-language-detection.

In order to load the dataset, run the following code:

from datasets import load_dataset

dataset = load_dataset("HiTZ/laion-eus")