aleizb/IDFACEIQA

IDCFIQA是一个身份一致性人脸图像生成质量评估基准数据集，旨在解决身份一致性人脸图像生成领域缺乏专门质量评估基准的问题。该数据集包含1,600张人脸图像，由八种代表性生成算法基于200张源人脸图像（其中100张为真实图像，100张为合成图像）生成。为了进行全面评估，数据集采用Bradley-Terry模型进行主观标注，涵盖两个关键维度：生成图像的感知质量和相对于源图像的身份一致性。此外，基于该基准，数据集还评估了多种现有图像质量评估方法，并深入分析了它们在视觉质量和身份一致性评估方面的能力。

IDCFIQA is an ID-consistent face image generation quality assessment benchmark dataset designed to address the lack of specialized quality assessment benchmarks in the field of ID-consistent face image generation. The dataset comprises 1,600 face images generated by eight representative generation algorithms, conditioned on 200 source face images (100 real and 100 synthetic). To provide a comprehensive evaluation, subjective annotations are conducted using the Bradley-Terry model across two essential dimensions: the perceptual quality of the generated images and their ID-consistency relative to the source images. Furthermore, leveraging this benchmark, a wide range of existing image quality assessment methods are evaluated, and an in-depth analysis of their capability to accurately assess both visual quality and identity consistency is provided.