README.md

metadata

dataset_info:
  features:
    - name: id
      dtype: int64
    - name: image_id
      dtype: string
    - name: eng
      dtype: string
    - name: afr
      dtype: string
    - name: amh
      dtype: string
    - name: bem
      dtype: string
    - name: cjk
      dtype: string
    - name: dik
      dtype: string
    - name: dyu
      dtype: string
    - name: ewe
      dtype: string
    - name: fuv
      dtype: string
    - name: hau
      dtype: string
    - name: ibo
      dtype: string
    - name: kik
      dtype: string
    - name: kab
      dtype: string
    - name: kam
      dtype: string
    - name: kon
      dtype: string
    - name: kmb
      dtype: string
    - name: lua
      dtype: string
    - name: lug
      dtype: string
    - name: lin
      dtype: string
    - name: kin
      dtype: string
    - name: yor
      dtype: string
  splits:
    - name: train
      num_bytes: 12340971
      num_examples: 8091
  download_size: 5936673
  dataset_size: 12340971
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: apache-2.0
task_categories:
  - translation

AfriMMD - African Multilingual Multimodal Dataset (POC)

AfriMMD is a multilingual dataset created to enhance linguistic diversity in AI, focusing on African languages. This is a proof-of-concept experiment on the use of multimodal datasets to represent African languages in AI. The dataset contains translations of the captions in the widely-used Flickr8k dataset into 20 African languages. The goal is to address the underrepresentation of African languages in AI and foster more inclusive AI technologies. The image-text pairs have been carefully translated into multiple African languages, providing an avenue for advanced and inclusive AI development, particularly in multimodal tasks that involve both text and images.

Images associated with the dataset can manually be downloaded from Github or Kaggle

Supported Languages

Amharic (amh), Bemba (bem), Chokwe (cjk), Rek (dik), Dyula (dyu), Ewe (ewe), Fulfulde (fuv), Hausa (hau), Igbo (ibo), Kikuyu (kik), Kabyle (kab), Kamba (kam), Kikongo (kon), Kimbundu (kmb), LubaKasai (lua), Ganda (lug), Lingala (lin), Kinyarwanda (kin), Yoruba (yor)

Load Dataset

from datasets import load_dataset

dataset = load_dataset('AfriMM/AfriMMD')

Applications

• Multilingual multimodal tasks (eg: image captioning in African languages, pre-trained vision-language models, etc.)
• Translation and language learning for supported African languages.
• Research on cross-cultural understanding and representation in AI.

Citation

@dataset{afrimm2024,
  author       = {AfriMM - ML Collective},
  title        = {AfriMMD: Multimodal Dataset for African Languages},
  year         = 2024,
  url          = {https://huggingface.co/datasets/AfriMM/AfriMMD}
}