CHAIRS
Emerging19papers using it
2024first seen
The 'CHAIRS' dataset is used to evaluate hallucination in vision-language models by measuring the accuracy of generated responses in relation to visual content.
Papers using CHAIRS (19)
- The Truth Stays in the Family: Enhancing Contextual Grounding via Inherited Truthful Heads in Model LineagesMitigating Hallucinations In Multimodal Llms Via Object-aware Preference OptimizationAgilePruner: An Empirical Study of Attention and Diversity for Adaptive Visual Token Pruning in Large Vision-Language ModelsMitigating Object Hallucinations in LVLMs via Attention Imbalance RectificationBeyond Dominant Patches: Spatial Credit Redistribution For Grounded Vision-Language ModelsHulluEdit: Single-Pass Evidence-Consistent Subspace Editing for Mitigating Hallucinations in Large Vision-Language ModelsSAVE: Sparse Autoencoder-driven Visual Information Enhancement For Mitigating Object HallucinationContext-Aware Decoding for Faithful Vision-Language GenerationAttention-space Contrastive Guidance for Efficient Hallucination Mitigation in LVLMsConscious Gaze: Adaptive Attention Mechanisms for Hallucination Mitigation in Vision-Language ModelsCausally-Grounded Dual-Path Attention Intervention for Object Hallucination Mitigation in LVLMsMaskCD: Mitigating LVLM Hallucinations by Image Head Masked Contrastive DecodingExposing Hallucinations To Suppress Them: VLMs Representation Editing With Generative AnchorsASCD: Attention-Steerable Contrastive Decoding for Reducing Hallucination in MLLMBIMA: Bijective Maximum Likelihood Learning Approach To Hallucination Prediction And Mitigation In Large Vision-language ModelsMitigating Hallucination in Large Vision-Language Models via Adaptive Attention CalibrationTARAC: Mitigating Hallucination in LVLMs via Temporal Attention Real-time Accumulative ConnectionESREAL: Exploiting Semantic Reconstruction to Mitigate Hallucinations in
Vision-Language ModelsEnhancing Vision-Language Model Reliability with Uncertainty-Guided Dropout Decoding