XIMAGENET-12

XIMAGENET-12 is an explainable visual benchmark dataset for model robustness evaluation. It consists of over 200K images with 15,410 manual semantic annotations. The dataset is designed to simulate real-world situations and evaluate the robustness of visual models.