CVPR2026 Hallucination Detection AI paper notes paper summaries Multimodal/VLM Reasoning Adversarial Robustness Diffusion Models LLM Alignment/RLHF

👻 Hallucination Detection¶

📷 CVPR2026 · 18 paper notes

📌 Same area in other venues: 🧪 ICML2026 (19) · 💬 ACL2026 (27) · 🔬 ICLR2026 (9) · 🤖 AAAI2026 (15) · 🧠 NeurIPS2025 (17) · 📹 ICCV2025 (4)

🔥 Top topics: Multimodal/VLM ×8 · Reasoning ×2

Beyond the Global Scores: Fine-Grained Token Grounding as a Robust Detector of LVLM Hallucinations: A patch-level LVLM hallucination detection framework is proposed. Hallucinated tokens are found to exhibit two characteristic signatures—dispersed attention patterns and low semantic alignment—based on which two lightweight metrics are designed: Attention Dispersion Score (ADS) and Cross-modal Grounding Consistency (CGC), achieving 90% detection accuracy.
Fighting Hallucinations with Counterfactuals: Diffusion-Guided Perturbations for LVLM Hallucination Suppression: This paper proposes CIPHER, a training-free test-time hallucination suppression method. In the offline phase, a diffusion model is used to generate counterfactual images, constructing the OHC-25K dataset, from which visual hallucination subspaces are extracted via SVD. During inference, hidden states are projected onto the orthogonal complement of this subspace, significantly reducing visual hallucinations in LVLMs without modifying model parameters or incurring additional inference overhead.
FINER: MLLMs Hallucinate under Fine-grained Negative Queries: This paper identifies that MLLMs suffer a dramatic increase in hallucination rates under fine-grained negative queries (queries involving multiple objects/attributes/relations with only one subtle error), proposes the FINER benchmark and FINER-Tuning (based on DPO), achieving up to 24.2% improvement on InternVL3.5-14B.
HulluEdit: Single-Pass Evidence-Consistent Subspace Editing for Mitigating Hallucinations in Large Vision-Language Models: This paper proposes HulluEdit, a single-pass, reference-free subspace editing framework that decomposes hidden states into three orthogonal subspaces—a visual evidence subspace, a conflicting prior subspace, and a residual uncertainty subspace—to selectively suppress hallucination patterns without interfering with visual grounding, achieving state-of-the-art hallucination mitigation on the POPE and CHAIR benchmarks.
HulluEdit: Single-Pass Evidence-Consistent Subspace Editing for Mitigating Hallucinations in LVLMs: This paper proposes HulluEdit, a single-pass, reference-model-free hallucination mitigation framework that orthogonally decomposes hidden states into a visual evidence subspace, a conflicting prior subspace, and a residual uncertainty subspace, selectively suppressing hallucination patterns without interfering with visual grounding, achieving state-of-the-art performance on POPE and CHAIR.
KVSmooth: Mitigating Hallucination in Multi-modal Large Language Models through Key-Value Smoothing: This paper proposes KVSmooth, a training-free plug-and-play inference-time method that applies adaptive exponential moving average (EMA) smoothing to KV-Cache guided by attention row entropy, effectively suppressing semantic drift and hallucination generation caused by sink tokens during decoding in multimodal large language models (MLLMs). On LLaVA-1.5, CHAIR_S is reduced from 41.8 to 18.2 (a 56% reduction), while F1 improves from 77.5 to 79.2.
Locate-then-Sparsify: Attribution Guided Sparse Strategy for Visual Hallucination Mitigation: This paper proposes LTS-FS (Locate-Then-Sparsify for Feature Steering), a framework that employs causal intervention-based attribution to identify hallucination-relevant layers and applies layer-wise sparse control over feature steering intensity according to attribution scores, effectively mitigating hallucinations in LVLMs while preserving generalization capability.
Mitigating Multimodal Hallucinations via Gradient-based Self-Reflection: This paper proposes GACD (Gradient-based Influence-Aware Constrained Decoding), which employs first-order Taylor gradient estimation to quantify each token's influence on the output. GACD simultaneously mitigates multimodal hallucinations caused by text-visual bias and co-occurrence bias at inference time, requiring neither auxiliary models nor fine-tuning.
Mitigating Object Hallucination in LVLMs via Attention Imbalance Rectification: This paper introduces the concept of Attention Imbalance to explain object hallucination in LVLMs, and proposes a lightweight decoding-time intervention method, AIR, which rectifies attention imbalance via cross-modal attention reallocation and variance-constrained projection regularization. AIR reduces hallucination rates by up to 35.1% and improves general capability by up to 15.9% across four LVLMs.
MoD-DPO: Towards Mitigating Cross-modal Hallucinations in Omni LLMs using Modality Decoupled Preference Optimization: This paper proposes MoD-DPO (Modality-Decoupled DPO), which decouples the contribution of each modality in multimodal LLMs via three mechanisms—invariance regularization, sensitivity regularization, and language-prior debiasing—to effectively mitigate cross-modal hallucinations (e.g., answering visual questions using auditory information). A closed-form optimal policy is also derived.
Overthinking Causes Hallucination: Tracing Confounder Propagation in Vision Language Models: This paper reveals a novel mechanism underlying VLM hallucinations — overthinking: the model generates an excessive number of competing object hypotheses in intermediate decoding layers, and confounders propagate across layers to corrupt the final prediction. The paper proposes the Overthinking Score to quantify inter-layer hypothesis diversity × uncertainty, achieving F1 of 78.9% on MSCOCO and 71.58% on the OOD benchmark AMBER.
Reallocating Attention Across Layers to Reduce Multimodal Hallucination: A lightweight, training-free plugin method is proposed to mitigate hallucination in Multimodal Large Reasoning Models (MLRMs) by identifying perceptual and reasoning attention heads and applying Class-Conditioned Rescaling to rebalance cross-layer attention distribution. The method achieves an average improvement of 4.2% across 5 benchmarks with negligible additional inference overhead.
Residual Decoding: Mitigating Hallucinations in Large Vision-Language Models via History-Aware Residual Guidance: This paper proposes Residual Decoding (ResDec), a training-free plug-and-play decoding strategy that identifies the semantic anchoring phase by analyzing U-shaped JSD patterns in historical token logit distributions, aggregates logits from this phase as residual guidance to steer current decoding, and effectively suppresses language-prior hallucinations in LVLMs at near-zero additional inference overhead.
Tell Model Where to Look: Mitigating Hallucinations in MLLMs by Vision-Guided Attention: This paper proposes Vision-Guided Attention (VGA), a training-free method that constructs precise visual grounding from the semantic features of visual tokens to guide model attention toward relevant visual regions, effectively mitigating hallucinations in MLLMs while remaining compatible with FlashAttention.
TriDF: Evaluating Perception, Detection, and Hallucination for Interpretable DeepFake Detection: This paper proposes TriDF — the first benchmark that comprehensively evaluates interpretable DeepFake detection across three dimensions: Perception, Detection, and Hallucination. It comprises 55K high-quality samples covering 16 DeepFake types and 3 modalities, and reveals a triadic coupling relationship in which accurate perception is a prerequisite for reliable detection, yet hallucination can severely undermine decision-making.
Understanding and Mitigating Hallucinations in Multimodal Chain-of-Thought Models: This paper systematically analyzes the causes of hallucinations in multimodal CoT models, identifies "divergent thinking" (associative reasoning) as the core trigger, and proposes a training-free detection and decoding intervention strategy based on visual entropy. The method reduces CHAIRS by over 30% on Object HalBench while maintaining or improving general reasoning capability.
Understanding the Role of Hallucination in Reinforcement Post-Training of Multimodal Reasoning Models: This paper proposes the Hallucination-as-Cue analytical framework, systematically investigating the true mechanisms underlying RL post-training of multimodal reasoning models via three modality-specific corruption strategies (blank image, random image, text removal). The study finds that GRPO training with 100% corrupted visual inputs still yields significant improvements in reasoning performance, challenging the prevailing assumption that RL training effectively leverages visual information.
Zina: Multimodal Fine-grained Hallucination Detection and Editing: Zina formalizes the task of multimodal fine-grained hallucination detection and editing, proposes a two-stage system (detector MLLM + reviewer MLLM) that delegates token copying to a deterministic function to reduce model burden, constructs the VisionHall dataset (6.9K human-annotated + 20K graph-based synthetic samples), and surpasses GPT-4o by 15.8 points in detection F1.