AI-generated marketing images (>10K) + human ratings for quality and realism
Creators:
Jochen Hartmann, Yannick Exner
Publication Date:
not specified
Data Category:
Dataset Description:
The GenImageNet dataset is a comprehensive collection of AI-generated marketing images accompanied by human evaluations assessing their quality and realism. It encompasses 10,320 synthetic marketing images created using seven state-of-the-art generative text-to-image models, including DALL-E 3, Midjourney v6, Firefly 2, Imagen 2, Imagine, Realistic Vision, and Stable Diffusion XL Turbo. Each image was generated based on prompts derived from 2,400 real-world, human-made images, facilitating a direct comparison between AI-generated and human-made marketing visuals. The dataset's total size is approximately 3.1 GB, making it substantial yet manageable for analysis. It comprises 10,320 observations, each corresponding to an individual AI-generated image. The temporal coverage of the dataset aligns with the period during which the images were generated and evaluated, specifically around July 2023. Structurally, the dataset is organized to include each AI-generated image alongside its corresponding human-made image prompt and the human evaluations. The human evaluations consist of 254,400 individual assessments, with each image receiving multiple ratings to ensure reliability. These evaluations cover aspects such as quality, realism, aesthetics, creativity, adherence to the prompt, and overall effectiveness in a marketing context.
Variables:
Details: