CCTVBench exposes a large gap between standard QA accuracy and contrastive consistency in traffic video reasoning for multimodal LLMs and introduces C-TCD to narrow that gap.
Title resolution pending
30 Pith papers cite this work, alongside 362 external citations. Polarity classification is still indexing.
30
Pith papers citing it
362
external citations · OpenAlex
citation-role summary
background 2
dataset 2
citation-polarity summary
representative citing papers
MoHallBench is a new benchmark evaluating motion hallucination in VideoLLMs from co-occurrence priors, sequential inference, and similarity confusion, revealing decoupling from action recognition performance.
The paper proposes an operator-level visual-token skipping framework for MLLMs that reduces TFLOPs by 33.7% on Qwen3-VL while retaining 99.5% performance across VQA benchmarks.
S³E framework finds excess decision-state displacement under semantic stress in multimodal models despite consistent correct forced-choice behavior.
CAS mitigates object hallucinations in MLLMs by extracting two context preference vectors from designed conflict samples and applying signed residual injection at mid-early MLP layers without retraining or added latency.
Visual CoT agents exhibit tool-use collapse where tool usage declines but task accuracy rises, and adding entropy regularization for rollout diversity produces the strongest performance.
JMed48k is a new benchmark of Japanese healthcare licensing exams used to evaluate 21 VLMs, with a paired image-removal audit revealing large differences in how models and professions benefit from visual content.
BICR trains a lightweight probe on contrastive hidden states from real versus blind images to detect visual grounding in LVLM predictions, outperforming baselines on calibration and discrimination with fewer parameters.
MIRL uses mutual information to guide trajectory selection and provide separate rewards for visual perception in RLVR for VLMs, achieving 70.22% average accuracy with 25% fewer full trajectories.
DetailVerifyBench supplies 1,000 images and densely annotated long captions to evaluate precise hallucination localization in multimodal large language models.
Experiments across code LLMs show no-review collapses fastest, human-gated filters slow collapse, and AI self-gates lose effect over time, degenerating to ungated self-training under self-confirming acceptance as proven via gated distributional reweighting and spectral analysis.
QK Product Steering suppresses dominant singular modes in the per-head QK product of selected middle layers via a closed-form query-only update, yielding 4.0% average relative CHAIR_s reduction on three GQA VLMs.
TGV-KV uses text-vision budgeting, weighted ranking, and prioritised retention to evict KV cache in VLMs while retaining 99.2% accuracy at 5% budget on VizWiz-VQA.
AsyMoE adds hyperbolic geometry for cross-modal hierarchies and evidence-priority experts to address vision-language asymmetry in LVLMs, reporting 1.5% average gains and 25.45% fewer active parameters.
POW3R adapts rubric criterion weights via rollout contrast in RLVR to improve mean reward, strict completion rates, and training speed over static rubric aggregation on multimodal and text tasks.
Deep Pre-Alignment uses a small VLM perceiver instead of ViT to pre-align visual features with LLM text space, yielding 1.9-3.0 point gains on multimodal benchmarks and 32.9% less language forgetting.
F^3A is a training-free visual token pruning router that treats pruning as task-conditioned evidence search and allocates a fixed vision token budget using question cues and frozen sparse heads without extra LLM passes.
Hallucinations in LVLMs largely arise from textual priors in prompts, and can be reduced by fine-tuning with preference optimization on grounded vs. hallucinated response pairs.
MuRF fuses multi-resolution features from frozen vision foundation models at inference time to create stronger representations without any training.
MLLMs exhibit a consistent recognition-reasoning inversion on discrete visual symbols across domains, underperforming on elementary perception while appearing competent on higher-level reasoning via linguistic compensation.
FineGen uses a VLM multi-agent pipeline to build FineGen-100K, a 147k-sample hierarchical dataset of attribute-specific hard negatives, reporting 96.7% validity and +14.4% downstream accuracy gain on hard samples in FG-OVD.
VLMs keep linear visual quantity codes and comparative magnitude sense into extrapolation, but fail at symbolic mapping, supporting a fractured magnitude hypothesis.
VCap pairs reference captions as witnesses with visual signals as adjudicators to deliver hypergeometric-precision rewards for RL in visual captioning, enabling an 8B model to outperform SOTA on benchmarks and improve weak-to-strong generalization.
RAC adds ranking-aware group loss and clean-corrupted pairwise loss to RL post-training to boost both accuracy and calibration in multimodal reasoning without extra annotations.
citing papers explorer
-
CCTVBench: Contrastive Consistency Traffic VideoQA Benchmark for Multimodal LLMs
CCTVBench exposes a large gap between standard QA accuracy and contrastive consistency in traffic video reasoning for multimodal LLMs and introduces C-TCD to narrow that gap.
-
MoHallBench: A Benchmark for Motion Hallucination in Video Large Language Models
MoHallBench is a new benchmark evaluating motion hallucination in VideoLLMs from co-occurrence priors, sequential inference, and similarity confusion, revealing decoupling from action recognition performance.
-
Attend, Transform, or Silence: Operator-Level Visual Skipping for Efficient Multimodal LLM Inference
The paper proposes an operator-level visual-token skipping framework for MLLMs that reduces TFLOPs by 33.7% on Qwen3-VL while retaining 99.5% performance across VQA benchmarks.
-
When Correct Decisions Hide Internal Stress: Decision-State Probing in Multimodal Language Models
S³E framework finds excess decision-state displacement under semantic stress in multimodal models despite consistent correct forced-choice behavior.
-
Rethinking Visual Neglect: Steering via Context-Preference for MLLM Hallucination Mitigation
CAS mitigates object hallucinations in MLLMs by extracting two context preference vectors from designed conflict samples and applying signed residual injection at mid-early MLP layers without retraining or added latency.
-
Diversity Over Frequency: Rethinking Tool Use in Visual Chain-of-Thought Agents
Visual CoT agents exhibit tool-use collapse where tool usage declines but task accuracy rises, and adding entropy regularization for rollout diversity produces the strongest performance.
-
JMed48k: A Multi-Profession Japanese Medical Licensing Benchmark for Vision-Language Model Evaluation
JMed48k is a new benchmark of Japanese healthcare licensing exams used to evaluate 21 VLMs, with a paired image-removal audit revealing large differences in how models and professions benefit from visual content.
-
Grounded or Guessing? LVLM Confidence Estimation via Blind-Image Contrastive Ranking
BICR trains a lightweight probe on contrastive hidden states from real versus blind images to detect visual grounding in LVLM predictions, outperforming baselines on calibration and discrimination with fewer parameters.
-
MIRL: Mutual Information-Guided Reinforcement Learning for Vision-Language Models
MIRL uses mutual information to guide trajectory selection and provide separate rewards for visual perception in RLVR for VLMs, achieving 70.22% average accuracy with 25% fewer full trajectories.
-
DetailVerifyBench: A Benchmark for Dense Hallucination Localization in Long Image Captions
DetailVerifyBench supplies 1,000 images and densely annotated long captions to evaluate precise hallucination localization in multimodal large language models.
-
When AI Reviews Its Own Code: Recursive Self-Training Collapse in Code LLMs
Experiments across code LLMs show no-review collapses fastest, human-gated filters slow collapse, and AI self-gates lose effect over time, degenerating to ungated self-training under self-confirming acceptance as proven via gated distributional reweighting and spectral analysis.
-
Spectral Query-Key Product Weight Steering for Training-Free VLM Hallucination Mitigation
QK Product Steering suppresses dominant singular modes in the per-head QK product of selected middle layers via a closed-form query-only update, yielding 4.0% average relative CHAIR_s reduction on three GQA VLMs.
-
TGV-KV: Text-Grounded KV Eviction for Vision-Language Models
TGV-KV uses text-vision budgeting, weighted ranking, and prioritised retention to evict KV cache in VLMs while retaining 99.2% accuracy at 5% budget on VizWiz-VQA.
-
Hyperbolic and Evidence-Prioritized Experts for Large Vision-Language Models
AsyMoE adds hyperbolic geometry for cross-modal hierarchies and evidence-priority experts to address vision-language asymmetry in LVLMs, reporting 1.5% average gains and 25.45% fewer active parameters.
-
Not Every Rubric Teaches Equally: Policy-Aware Rubric Rewards for RLVR
POW3R adapts rubric criterion weights via rollout contrast in RLVR to improve mean reward, strict completion rates, and training speed over static rubric aggregation on multimodal and text tasks.
-
Deep Pre-Alignment for VLMs
Deep Pre-Alignment uses a small VLM perceiver instead of ViT to pre-align visual features with LLM text space, yielding 1.9-3.0 point gains on multimodal benchmarks and 32.9% less language forgetting.
-
How Many Visual Tokens Do Multimodal Language Models Need? Scaling Visual Token Pruning with F^3A
F^3A is a training-free visual token pruning router that treats pruning as task-conditioned evidence search and allocates a fixed vision token budget using question cues and frozen sparse heads without extra LLM passes.
-
When Prompts Override Vision: Prompt-Induced Hallucinations in LVLMs
Hallucinations in LVLMs largely arise from textual priors in prompts, and can be reduced by fine-tuning with preference optimization on grounded vs. hallucinated response pairs.
-
MuRF: Unlocking the Multi-Scale Potential of Vision Foundation Models
MuRF fuses multi-resolution features from frozen vision foundation models at inference time to create stronger representations without any training.
-
Cognitive Mismatch in Multimodal Large Language Models for Discrete Symbol Understanding
MLLMs exhibit a consistent recognition-reasoning inversion on discrete visual symbols across domains, underperforming on elementary perception while appearing competent on higher-level reasoning via linguistic compensation.
-
FineGen: A VLM-based Multi-Agent Framework for Fine-Grained Image-Text Dataset Construction
FineGen uses a VLM multi-agent pipeline to build FineGen-100K, a 147k-sample hierarchical dataset of attribute-specific hard negatives, reporting 96.7% validity and +14.4% downstream accuracy gain on hard samples in FG-OVD.
-
Unveiling the Visual Counting Bottleneck in Vision-Language Models
VLMs keep linear visual quantity codes and comparative magnitude sense into extrapolation, but fail at symbolic mapping, supporting a fractured magnitude hypothesis.
-
VCap: Hypergeometric Rewards for Weak-to-Strong Visual Captioning
VCap pairs reference captions as witnesses with visual signals as adjudicators to deliver hypergeometric-precision rewards for RL in visual captioning, enabling an 8B model to outperform SOTA on benchmarks and improve weak-to-strong generalization.
-
Ranking-Aware Calibration for Reliable Multimodal Reinforcement Learning
RAC adds ranking-aware group loss and clean-corrupted pairwise loss to RL post-training to boost both accuracy and calibration in multimodal reasoning without extra annotations.
-
MHSA: A Lightweight Framework for Mitigating Hallucinations via Steered Attention in LVLMs
MHSA mitigates hallucinations in LVLMs by training an MLP to steer cross-modal attention, extending detection work to mitigation via attention replacement at inference.
-
Not Blind but Silenced: Rebalancing Vision and Language via Adversarial Counter-Commonsense Equilibrium
ACE uses adversarial counter-commonsense perturbations on image tokens during decoding to suppress hallucinated linguistic priors while preserving stable visual signals in MLLMs.
-
Combating Textual Noise and Redundancy: Entropy-Aware Dense Visual Token Pruning
EADP filters textual noise via statistical entropy then casts token selection as submodular maximization with spatial prior to preserve fine-grained cues in VLMs under strict budgets.
-
Transparent and Controllable Recommendation Filtering via Multimodal Multi-Agent Collaboration
A multi-agent multimodal system with fact-grounded adjudication and a dynamic two-tier preference graph cuts false positives in content filtering by 74.3% and nearly doubles F1-score versus text-only baselines while supporting user-driven Delta adjustments.
-
Parallel LLM Reasoning for Bias-Resilient, Robust Conceptual Abstraction
Parallel chunk processing with evidence-anchored consolidation reduces omission errors by 84%, boosts traceability by 130%, and cuts unsupported claims by 91% in LLM long-document analysis.
-
A Survey on LLM-as-a-Judge
A survey on LLM-as-a-Judge that reviews reliability strategies, proposes evaluation methods, and introduces a novel benchmark for assessing such systems.