During compilation, the Preprocessor processes the source code (SRC) to eliminate comments and expand macros or includes. The cleaned code is then forwarded to the Compiler, which converts it into ...

Abstract: Vision-language pre-training models have demonstrated outstanding performance on a wide range of multimodal tasks. Nevertheless, they remain susceptible to multimodal adversarial examples.