SafeAtlas-VL: Beyond Binary Multimodal Safety with Large-Scale Data and Guard Models
This paper introduces SafeAtlas-VL, a dataset of 1.5M training instances that places image-, request-, and response-level judgments on a five-level ordered scale, and trains the SafeAtlas Guard series of models via target-conditioned tuning for multimodal safety detection.
Zong-Rui Wang, Xiang-Yang Zhu, Sixiang Wang et al.
· 0 citations