# SenBen: Sensitive Scene Graphs for Explainable Content Moderation

arXiv: https://arxiv.org/abs/2604.08819 · HTML: https://mlsanity.com/paper/2604.08819/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 26 of its 40 references that are in the map.
- First posted on arXiv 9 Apr 2026.

## Builds on (most cited first, 26 of 26)

- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [Scheduled Sampling for Sequence Prediction with Recurrent Neural Networks](https://mlsanity.com/paper/1506.03099/) (2015)
- [Neural Motifs: Scene Graph Parsing with Global Context](https://mlsanity.com/paper/1711.06640/) (2017)
- [Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks](https://mlsanity.com/paper/2311.06242/) (2023)
- [Asymmetric Loss For Multi-Label Classification](https://mlsanity.com/paper/2009.14119/) (2020)
- [Unbiased Scene Graph Generation from Biased Training](https://mlsanity.com/paper/2002.11949/) (2020)
- [Pix2seq: A Language Modeling Framework for Object Detection](https://mlsanity.com/paper/2109.10852/) (2021)
- [MovieNet: A Holistic Dataset for Movie Understanding](https://mlsanity.com/paper/2007.10937/) (2020)
- [Movie Description](https://mlsanity.com/paper/1605.03705/) (2016)
- [Query2Label: A Simple Transformer Way to Multi-Label Classification](https://mlsanity.com/paper/2107.10834/) (2021)
- [ABAW: Valence-Arousal Estimation, Expression Recognition, Action Unit Detection & Emotional Reaction Intensity Estimation Challenges](https://mlsanity.com/paper/2303.01498/) (2023)
- [Distil-Whisper: Robust Knowledge Distillation via Large-Scale Pseudo Labelling](https://mlsanity.com/paper/2311.00430/) (2023)
- [MovieGraphs: Towards Understanding Human-Centric Situations from Videos](https://mlsanity.com/paper/1712.06761/) (2017)
- [Order-Agnostic Cross Entropy for Non-Autoregressive Machine Translation](https://mlsanity.com/paper/2106.05093/) (2021)
- [UnsafeBench: Benchmarking Image Safety Classifiers on Real-World and AI-Generated Images](https://mlsanity.com/paper/2405.03486/) (2024)
- [LlavaGuard: An Open VLM-based Framework for Safeguarding Vision Datasets and Models](https://mlsanity.com/paper/2406.05113/) (2024)
- [FACTUAL: A Benchmark for Faithful and Consistent Textual Scene Graph Parsing](https://mlsanity.com/paper/2305.17497/) (2023)
- [Skeleton Recall Loss for Connectivity Conserving and Resource Efficient Segmentation of Thin Tubular Structures](https://mlsanity.com/paper/2404.03010/) (2024)
- [ShieldGemma 2: Robust and Tractable Image Content Moderation](https://mlsanity.com/paper/2504.01081/) (2025)
- [Fine-Grained Predicates Learning for Scene Graph Generation](https://mlsanity.com/paper/2204.02597/) (2022)
- [σ-GPTs: A New Approach to Autoregressive Models](https://mlsanity.com/paper/2404.09562/) (2024)
- [Striking the Right Balance: Recall Loss for Semantic Segmentation](https://mlsanity.com/paper/2106.14917/) (2021)
- [Compile Scene Graphs with Reinforcement Learning](https://mlsanity.com/paper/2504.13617/) (2025)
- [Towards Unbiased and Robust Spatio-Temporal Scene Graph Generation and Anticipation](https://mlsanity.com/paper/2411.13059/) (2024)
- [RA-SGG: Retrieval-Augmented Scene Graph Generation Framework via Multi-Prototype Learning](https://mlsanity.com/paper/2412.12788/) (2024)
- [State-of-the-Art in Nudity Classification: A Comparative Analysis](https://mlsanity.com/paper/2312.16338/) (2023)

## Abstract (arXiv)

Content moderation systems classify images as safe or unsafe but lack spatial grounding and interpretability: they cannot explain what sensitive behavior was detected, who is involved, or where it occurs. We introduce the Sensitive Benchmark (SenBen), the first large-scale scene graph benchmark for sensitive content, comprising 13,999 frames from 157 movies annotated with Visual Genome-style scene graphs (25 object classes, 28 attributes including affective states such as pain, fear, aggression, and distress, 14 predicates) and 16 sensitivity tags across 5 categories. We distill a frontier VLM into a compact 241M student model using a multi-task recipe that addresses vocabulary imbalance in autoregressive scene graph generation through suffix-based object identity, Vocabulary-Aware Recall (VAR) Loss, and a decoupled Query2Label tag head with asymmetric loss, yielding a +6.4 percentage point improvement in SenBen Recall over standard cross-entropy training. On grounded scene graph metrics, our student model outperforms all evaluated VLMs except Gemini models and all commercial safety APIs, while achieving the highest object detection and captioning scores across all models, at 7.6× faster inference and 16× less GPU memory.