README.md

metadata

license: mit
dataset_info:
  features:
    - name: image
      dtype: image
    - name: label
      dtype:
        class_label:
          names:
            '0': ai_generated
            '1': human_created
    - name: image_name
      dtype: string
  splits:
    - name: train
      num_bytes: 831331860
      num_examples: 11998
    - name: test
      num_bytes: 102610854
      num_examples: 1500
    - name: validation
      num_bytes: 101540782
      num_examples: 1499
  download_size: 1035437916
  dataset_size: 1035483496
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
      - split: validation
        path: data/validation-*
task_categories:
  - image-classification
language:
  - en
tags:
  - images
  - ai
pretty_name: Skynet Detection
size_categories:
  - 10K<n<100K

Description

Roughly 15k images evenly split between AI generated and human generated for the purpose of training models to detect AI generated content. Images have been converted to numpy arrays and resized to 224x224x3 for alignment with CNN models available on Tensorflow

Sources

The following sources were used to collect the images

AI vs Human
Open Images V7
Laion - 400M

Dataset_Info

Features:

• name: image
  dtype: numpy arrays (224x224x3)

• name: label
  dtype: class_label

  • names:
    • '0': ai_generated
    • '1': human_created

• name: image_name
  dtype: string
  description: unique index for tracking image

Splits:

• name: train
  num_bytes: 831 Mb
  num_examples: 11,998

• name: test
  num_bytes: 102 Mb
  num_examples: 1,500

• name: validation
  num_bytes: 101 Mb
  num_examples: 1,499

download_size: 1.04 Gb
dataset_size: 1.04 Gb

Configs:

• config_name: default

  data_files:

  • split: train
    path: data/train-*
  • split: test
    path: data/test-*
  • split: validation
    path: data/validation-*