Perceptual Artifacts Localization for Image Synthesis Tasks
This work addresses the need for automated correction of artifacts in image synthesis, which is incremental as it builds on existing generative models by providing localization and rectification tools.
The study tackled the problem of perceptual artifacts in generated images by creating a dataset of 10,168 images with per-pixel artifact labels across ten synthesis tasks and training a segmentation model to localize these artifacts, achieving effective adaptation to unseen models with minimal samples and proposing a zoom-in inpainting pipeline for rectification.
Recent advancements in deep generative models have facilitated the creation of photo-realistic images across various tasks. However, these generated images often exhibit perceptual artifacts in specific regions, necessitating manual correction. In this study, we present a comprehensive empirical examination of Perceptual Artifacts Localization (PAL) spanning diverse image synthesis endeavors. We introduce a novel dataset comprising 10,168 generated images, each annotated with per-pixel perceptual artifact labels across ten synthesis tasks. A segmentation model, trained on our proposed dataset, effectively localizes artifacts across a range of tasks. Additionally, we illustrate its proficiency in adapting to previously unseen models using minimal training samples. We further propose an innovative zoom-in inpainting pipeline that seamlessly rectifies perceptual artifacts in the generated images. Through our experimental analyses, we elucidate several practical downstream applications, such as automated artifact rectification, non-referential image quality evaluation, and abnormal region detection in images. The dataset and code are released.