Gen-100

Recent advances in generative visual content have led to a quantum leap in the quality of artificially generated Deepfake content. Especially, diffusion models are causing growing concerns among communities due to their ever-increasing realism. However, quantifying the realism of generated content is still challenging. Existing evaluation metrics, such as Inception Score and Fréchet inception distance, fall short on benchmarking diffusion models due to the versatility of the generated images. To address this, we propose the \textbf{I}mage \textbf{R}ealism \textbf{S}core (IRS) evaluation metric, computed from five statistical measures of a given image. This non-learning-based metric not only efficiently quantifies the realism of generated images, but it is also a viable tool for detecting if an image is real or fake. To facilitate further efforts towards the quantification of realism in diffusion-generated content, we also introduce a new dataset, Gen-100. It consists of 100 categories, each featuring 30 images produced using prompts from ChatGPT with various models, including Stable Diffusion Model (SDM),Dalle2, Midjourney, and BigGAN.

近期在生成式视觉内容领域取得的突破性进展，使得人工生成的深度伪造内容的质量实现了质的飞跃。尤其是扩散模型因其日益增强的现实感而引发了社区的广泛关注。然而，对生成内容的现实感进行量化仍然是一个挑战。现有的评估指标，如Inception Score和Fréchet inception distance，由于生成图像的多样性，在评估扩散模型时存在不足。为此，我们提出了 extbf{图像现实度评分}(IRS)这一评估指标，该指标基于给定图像的五个统计量计算得出。这一非学习型指标不仅能够有效地量化生成图像的现实感，而且成为了一种检测图像真实或伪造的可行工具。为了促进对扩散生成内容现实感量化的进一步研究，我们还引入了一个新的数据集，即Gen-100。该数据集包含100个类别，每个类别包含由ChatGPT的提示生成的30张图像，使用了包括稳定扩散模型（SDM）、Dalle2、Midjourney和BigGAN在内的多种模型。