The PRISM benchmark dataset supports research on detecting generated images. It contains photorealistic synthesized and manipulated images, with corresponding real images sourced from the COCO 2017 Train and Validation sets. The dataset was created by oppiliF and is associated with a 2026 research paper.
Use Cases
- Benchmarking detection models for synthesized images based on photorealistic image content
- Training classifiers to distinguish real from generated images based on the described image pairs
- Evaluating the robustness of image authenticity algorithms against advanced synthesis techniques
Strengths
- Real images are sourced from the established COCO 2017 dataset, providing a known reference
- Dataset is directly associated with a published research paper, providing documented context
Limitations
- Column-level documentation is absent; field semantics must be inferred after download
- Row count is unknown, which may limit suitability assessment
Provenance
- Source
- COCO 2017 dataset for real images; synthesized images from the PRISM benchmark.
- Freshness
- Last updated 2026-06-06 21:47:53; freshness should be verified