README.md

metadata

configs:
  - config_name: default
    data_files:
      - split: train
        path:
          - data/train-*
          - data/val-*
      - split: validation
        path: data/validation-*
dataset_info:
  features:
    - name: id
      dtype: string
    - name: original_prompt
      dtype: string
    - name: positive_prompt
      dtype: string
    - name: negative_prompt
      dtype: string
    - name: url
      dtype: string
    - name: model_gen0
      dtype: string
    - name: model_gen1
      dtype: string
    - name: model_gen2
      dtype: string
    - name: model_gen3
      dtype: string
    - name: width_gen0
      dtype: int64
    - name: width_gen1
      dtype: int64
    - name: width_gen2
      dtype: int64
    - name: width_gen3
      dtype: int64
    - name: height_gen0
      dtype: int64
    - name: height_gen1
      dtype: int64
    - name: height_gen2
      dtype: int64
    - name: height_gen3
      dtype: int64
    - name: num_inference_steps_gen0
      dtype: int64
    - name: num_inference_steps_gen1
      dtype: int64
    - name: num_inference_steps_gen2
      dtype: int64
    - name: num_inference_steps_gen3
      dtype: int64
    - name: filepath_gen0
      dtype: string
    - name: filepath_gen1
      dtype: string
    - name: filepath_gen2
      dtype: string
    - name: filepath_gen3
      dtype: string
    - name: image_gen0
      dtype: image
    - name: image_gen1
      dtype: image
    - name: image_gen2
      dtype: image
    - name: image_gen3
      dtype: image
  splits:
    - name: train
      num_bytes: 2626848010531.5
      num_examples: 2306629
    - name: validation
      num_bytes: 5318900038
      num_examples: 4800
  download_size: 2568003790242
  dataset_size: 2632166910569.5

ELSA - Multimedia use case

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

Dataset Summary

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. This dataset can be used to train robust and accurate models that can identify and flag instances of deep fake images.

ELSA versions

Name	Description	Link
ELSA1M_track1	Dataset of 1M images generated using diffusion model	https://huggingface.co/datasets/elsaEU/ELSA1M_track1
ELSA10M_track1	Dataset of 10M images generated using four different diffusion models for each caption, multiple image compression formats, multiple aspect ration	https://huggingface.co/datasets/elsaEU/ELSA_D3
ELSA500k_track2	Dataset of 500k images generated using diffusion model with diffusion attentive attribution maps [1]	https://huggingface.co/datasets/elsaEU/ELSA500k_track2

from datasets import load_dataset
elsa_data = load_dataset("elsaEU/ELSA_D3", split="train", streaming=True)

Using streaming=True lets you work with the dataset without downloading it.

Dataset Structure

Each parquet file contains nearly 1k images and a JSON file with metadata.

The Metadata for generated images are:

• ID: Laion image ID
• original_prompt: Laion Prompt
• positive_prompt: positive prompt used for image generation
• negative_prompt: negative prompt used for image generation
• url: Url of the real image associated with the same prompt
• width: width generated image
• height: height generated image
• num_inference_steps: diffusion steps of the generator
• filepath: path of the generated image
• model_gen0: Generator 0 name
• model_gen1: Generator 1 name
• model_gen2: Generator 2 name
• model_gen3: Generator 3 name
• image_gen0: image generated with generator 0
• image_gen1: image generated with generator 1
• image_gen2: image generated with generator 2
• image_gen3: image generated with generator 3
• aspect_ratio: aspect ratio of the generated image

Dataset Curators

• Leonardo Labs ([email protected])
• UNIMORE (https://aimagelab.ing.unimore.it/imagelab/)

Paper page

Paper can be found at https://huggingface.co/papers/2407.20337.