Percept-Lens

<p><strong>Percept-Lens Dataset Description</strong></p>

<p><strong>Purpose.</strong><br>
The Percept-Lens dataset is curated as part of the <em>Percept-Lens</em> benchmark to evaluate the generalization capability of AI-generated image detectors under prompt-induced distribution shifts. It serves as an out-of-distribution (OOD) test set, designed to probe model robustness against semantically rich prompts not seen during training.</p>

<p><strong>Nature.</strong><br>
This dataset contains files to create the dataset. It contains the parquet files containing the prompts for various datasets. These prompts span abstract concepts, visual reasoning, compositional scenes, and imaginative queries. Images were generated using diverse diffusion models, including <em>Stable Diffusion variants</em>, and <em>FLUX.variants</em>, ensuring stylistic variation. No real images are included—each image is the synthetic output of a generative model responding to a high-level prompt.</p>

<p><strong>Scope.</strong><br>
The dataset includes original prompt used in creating the Percept-Lens Dataset. It is used exclusively for evaluation within the Percept-Lens benchmark and is not part of the training data. This subset is particularly useful for testing <em>semantic generalization</em>, where linguistic complexity drives visual synthesis beyond familiar domains.