CHAIR

The 'CHAIR' dataset/benchmark is used to evaluate the performance of vision-language models by measuring their ability to generate accurate and coherent captions while minimizing hallucinations.

Papers using CHAIR (22)

The Truth Stays in the Family: Enhancing Contextual Grounding via Inherited Truthful Heads in Model Lineages2026Mitigating Hallucinations In Multimodal Llms Via Object-aware Preference Optimization2025When Language Overwrites Vision: Over-Alignment and Geometric Debiasing in Vision-Language Models2026Mitigating Object Hallucinations in Vision-Language Models through Region-Aware Attention Recalibration2026See Fair, Speak Truth: Equitable Attention Improves Grounding and Reduces Hallucination in Vision-Language Alignment2026Global Context or Local Detail? Adaptive Visual Grounding for Hallucination Mitigation2026AgilePruner: An Empirical Study of Attention and Diversity for Adaptive Visual Token Pruning in Large Vision-Language Models2026Mitigating Object Hallucinations in LVLMs via Attention Imbalance Rectification2026Beyond Dominant Patches: Spatial Credit Redistribution For Grounded Vision-Language Models2026HulluEdit: Single-Pass Evidence-Consistent Subspace Editing for Mitigating Hallucinations in Large Vision-Language Models2026Context-Aware Decoding for Faithful Vision-Language Generation2026Attention-space Contrastive Guidance for Efficient Hallucination Mitigation in LVLMs2026Conscious Gaze: Adaptive Attention Mechanisms for Hallucination Mitigation in Vision-Language Models2025Causally-Grounded Dual-Path Attention Intervention for Object Hallucination Mitigation in LVLMs2025MaskCD: Mitigating LVLM Hallucinations by Image Head Masked Contrastive Decoding2025Exposing Hallucinations To Suppress Them: VLMs Representation Editing With Generative Anchors2025ASCD: Attention-Steerable Contrastive Decoding for Reducing Hallucination in MLLM2025BIMA: Bijective Maximum Likelihood Learning Approach To Hallucination Prediction And Mitigation In Large Vision-language Models2025Mitigating Hallucination in Large Vision-Language Models via Adaptive Attention Calibration2025TARAC: Mitigating Hallucination in LVLMs via Temporal Attention Real-time Accumulative Connection2025ESREAL: Exploiting Semantic Reconstruction to Mitigate Hallucinations in Vision-Language Models2024Enhancing Vision-Language Model Reliability with Uncertainty-Guided Dropout Decoding2024