Search papers, labs, and topics across Lattice.
Tsinghua AICi锚ncia e Tecnologia de Minas GeraisPol铆cia Federal (PF)SheffieldUniversidade de S茫o Paulo (USP)Universidade Estadual de CampinasUniversidade Federal de Minas GeraisApr 8, 2026arXiv:2604.07132The paper introduces CSA-Graphs, a privacy-preserving structural dataset for Child Sexual Abuse Imagery (CSAI) classification, addressing the challenge of data scarcity due to legal and ethical restrictions. The dataset consists of scene graphs (object relationships) and skeleton graphs (human pose) derived from CSAI images, effectively removing explicit visual content while retaining contextual information. Experiments demonstrate that these graph-based representations, especially when combined, maintain sufficient information for CSAI classification, enabling broader research in this sensitive area.
Unlock progress in child safety AI: a new dataset lets researchers train models on the structure of abusive images, without ever seeing the original content.
Child Sexual Abuse Imagery (CSAI) classification is an important yet challenging problem for computer vision research due to the strict legal and ethical restrictions that prevent the public sharing of CSAI datasets. This limitation hinders reproducibility and slows progress in developing automated methods. In this work, we introduce CSA-Graphs, a privacy-preserving structural dataset. Instead of releasing the original images, we provide structural representations that remove explicit visual content while preserving contextual information. CSA-Graphs includes two complementary graph-based modalities: scene graphs describing object relationships and skeleton graphs encoding human pose. Experiments show that both representations retain useful information for classifying CSAI, and that combining them further improves performance. This dataset enables broader research on computer vision methods for child safety while respecting legal and ethical constraints.
