4 papers
Same Attention, Different Truths: Put Logit-Lens over Visual Attention to Detect and Mitigate LVLM Object Hallucination
Zichuan Wang, Songlin Yang, Bo Peng +4
Large Vision-Language Models (LVLMs) often suffer from object hallucination, generating objects that are absent from the image. Prior work largely attributes this to insufficient v…
Endogenous Reprompting: Self-Evolving Cognitive Alignment for Unified Multimodal Models
Zhenchen Tang, Songlin Yang, Zichuan Wang +4
Unified Multimodal Models (UMMs) exhibit strong understanding, yet this capability often fails to effectively guide generation. We identify this as a Cognitive Gap: the model lacks…
Instant Preference Alignment for Text-to-Image Diffusion Models
Yang Li, Songlin Yang, Xiaoxuan Han +4
Text-to-image (T2I) generation has greatly enhanced creative expression, yet achieving preference-aligned generation in a real-time and training-free manner remains challenging. Pr…
Probing Unlearned Diffusion Models: A Transferable Adversarial Attack Perspective
Xiaoxuan Han, Songlin Yang, Wei Wang +2
Advanced text-to-image diffusion models raise safety concerns regarding identity privacy violation, copyright infringement, and Not Safe For Work content generation. Towards this,…