T/CITIF 010-2023

Scope: This document specifies the robustness grading requirements and performance evaluation methods for artificial intelligence systems used to detect the safety of image content. This document is applicable to third-party inspection and testing institutions, technology manufacturers, and technology application parties for testing and evaluating the robustness of artificial intelligence systems for content safety detection. Main technical content: It specifies the robustness grading requirements for artificial intelligence systems used to detect the safety of image content. According to the generation method of test samples and the difficulty of data acquisition, the test samples are graded. Through the test results of test samples of different levels, comprehensive evaluation indicators, processes, and methods for evaluating the robustness performance of artificial intelligence systems are constructed.