Tag
This paper defines multi-image implicit toxicity (MIIT), where individually benign images become toxic when combined, and proposes MiShield, a model trained with progressively distilled reasoning supervision to detect MIIT. Experiments show MiShield-8B outperforms existing moderation services.
ReMMD introduces a realistic multilingual multi-image agentic verification framework for multimodal misinformation detection, including a benchmark (ReMMDBench) with 500 samples and 2,756 images, and an agent (ReMMD-Agent) that achieves superior veracity performance with reduced costs.
IndustryBench-MIPU is a large-scale benchmark for multi-image industrial product understanding, evaluating 9 MLLMs and revealing a completeness gap where precision is high but attribute recovery is low.
This paper introduces a stateful visual encoder for vision-language models that conditions visual representations on prior features, enabling better visual comparison in multi-image and agentic settings. The method shows consistent improvements across tasks such as cross-image spatial aggregation and longitudinal radiology.