ConImageGen

# ConImageGen **ConImageGen** is a benchmark for **cross-paradigm AI-generated image detection**, covering both: * **Image-free generation** * **Image-conditioned generation** The dataset contains **364K images**, including **182K AI-generated images** and **182K real images**, generated by **13 representative models**. ## Dataset ConImageGen contains: * 8 image-free generators * 5 image-conditioned generators * Train / validation / test splits * Baseline resources for evaluation The dataset is designed to evaluate both **cross-generator** and **cross-paradigm generalization** of AI-generated image detectors. ## Download ```bash pip install modelscope modelscope download \ --dataset ConImageGen/ConImageGen-dataset \ --local_dir ./ConImageGen ``` ## Code & Checkpoints The implementation and pretrained checkpoints of **DTS-Det** are available at: https://www.modelscope.cn/models/ConImageGen/ConImageGen-checkpoints ## Anonymous Artifact This dataset is released as an **anonymous artifact for peer review**. Author and affiliation information is intentionally omitted during the review process.