README.md

metadata

configs:
  - config_name: default
    data_files:
      - split: test_set
        path:
          - data/train-*
      - split: test_set_transf
        path:
          - data/test_*
dataset_info:
  features:
    - name: image
      dtype: image
    - name: id
      dtype: string
  splits:
    - name: test_set
      num_bytes: 60255209120.125
      num_examples: 97111
    - name: test_set_transf
      num_bytes: 49096232796.125
      num_examples: 97111
  download_size: 109356949734
  dataset_size: 109351441916.25

ELSA - Multimedia use case

from datasets import load_dataset
elsa_test_set = load_dataset("elsaEU/ELSA_D3", split="test_set", streaming=True)

elsa_test_set_transf = load_dataset("elsaEU/ELSA_D3", split="test_set_transf", streaming=True)

Official page of the D3 external test set, composed by two different splits.

ELSA Multimedia D3 is a large collection of Deep Fake images, generated using diffusion models

This dataset was developed as part of the EU project ELSA. Specifically for the Multimedia use-case. Official webpage: https://benchmarks.elsa-ai.eu/ This dataset aims to develop effective solutions for detecting and mitigating the spread of deep fake images in multimedia content. Deep fake images, which are highly realistic and deceptive manipulations, pose significant risks to privacy, security, and trust in digital media. The training set is reachable at this link: https://huggingface.co/datasets/elsaEU/ELSA_D3