AI Paper Atlas

Research map / Benchmarks / POPE Adversarial

POPE Adversarial leaderboard

41 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on POPE Adversarial. Best accuracy: 90.5 (Owl, AAAI 2026).

Numbers are as each paper reports them, read by AI from its results tables and not checked by hand. Papers can differ in setting (training data, model size, extra data); the topic column shows what each paper is about. Open the paper before you compare or quote a number.

POPE Adversarial · accuracy

29 papers, each with its best reported number for its own method; higher is better.

#accuracy ↑MethodPaperTopic
190.5Owl
LLaVA-1.5
Causally-Grounded Dual-Path Attention Intervention for Object Hallucination Mitigation in LVLMs
AAAI 2026 · Table 2
Multimodal LLMs
288.2VCGD
InternVL-26B
VCGD: Visual Clue Guided Decoding with Caption Model for Mitigating Hallucination in Multimodal Large Language Models
AAAI 2026 · Table 1
Multimodal LLMs
387.47CODE
Emu2-Chat 14B
CODE: Contrasting Self-generated Description to Combat Hallucination in Large Multi-modal Models
NeurIPS 2024 · Table 1
Multimodal LLMs
487.26PND
Qwen3VL-2B
Breaking the Illusion: When Positive Meets Negative in Multimodal Decoding
CVPR 2026 · Table 1
Multimodal LLMs
586.98CRG
Qwen2.5-VL-7B-Instruct
Mitigating Hallucinations in Large Vision-Language Models via Causal Route Gating
ICML 2026 · Table 1
Multimodal LLMs
686.33GROUNDHOG-7B
Vicuna-7B
GROUNDHOG: Grounding Large Language Models to Holistic Segmentation
CVPR 2024 · POPE results table
Multimodal LLMs
786.3LLaVA-1.5+TDSR
LLaVA-1.5-7B
Top-Down Semantic Refinement for Image Captioning
AAAI 2026 · text
Multimodal LLMs
886.1Qwen-VL VL-SAE
Qwen-VL
VL-SAE: Interpreting and Enhancing Vision-Language Alignment with a Unified Concept Set
NeurIPS 2025 · Table: Experimental results on
CLIP and zero-shot
985.67IRI
LLaVA-NeXT
Causal Tracing of Object Representations in Large Vision Language Models: Mechanistic Interpretability and Hallucination Mitigation
AAAI 2026 · Table 2
Multimodal LLMs
1085.67RGPT
LLaVA
RegionGPT: Towards Region Understanding Vision Language Model
CVPR 2024 · Table: Results on the object h
Multimodal LLMs
19 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

POPE Adversarial · F1

16 papers, each with its best reported number for its own method; higher is better.

#F1 ↑MethodPaperTopic
193.9Dehall
GPT-4V
Combating Multimodal LLM Hallucination via Bottom-Up Holistic Reasoning
AAAI 2025 · Table 2
Multimodal LLMs
289.8VCGD
InternVL-26B
VCGD: Visual Clue Guided Decoding with Caption Model for Mitigating Hallucination in Multimodal Large Language Models
AAAI 2026 · Table 1
Multimodal LLMs
389.8SFT + VIGIL
Qwen2.5-VL-72B
Staying VIGILant: Mitigating Visual Laziness in MLLMs via Information-Theoretic Alignment
ECCV 2026 · Table 1
Multimodal LLMs
488.47TAF
Qwen2.5-VL-7B
Taming the Phantom: Token-Asymmetric Filtering for Hallucination Mitigation in Large Vision-Language Models
AAAI 2026 · Table 1
Multimodal LLMs
587.63REVIS
Qwen2.5-VL-7B-Instruct
REVIS: Sparse Latent Steering to Mitigate Object Hallucination in Large Vision-Language Models
ICML 2026 · Table 2
Multimodal LLMs
685.9CogAgent
Vicuna-7B
CogAgent: A Visual Language Model for GUI Agents
CVPR 2024 · MM-Vet and POPE table
Multimodal LLMs
785.6ObjecTok-LLaVA
Vicuna-7B
ObjecTok: Learning Holistic and Robust Object Tokens for MLLMs
AAAI 2026 · Table 1
Multimodal LLMs
884.9HSA-DPO w/ LLaVA-1.5
LLaVA-1.5
Detecting and Mitigating Hallucination in Large Vision Language Models via Fine-Grained AI Feedback
AAAI 2025 · Table 3
Multimodal LLMs
984.71LIONLION: Empowering Multimodal Large Language Model with Dual-Level Visual Knowledge
CVPR 2024 · POPE table
Multimodal LLMs
1084.35VisFlow
LLaVA-1.5
Not All Tokens and Heads Are Equally Important: Dual-Level Attention Intervention for Hallucination Mitigation
AAAI 2026 · Table 1
Multimodal LLMs
6 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

Not affiliated with any conference, journal or the benchmark's authors. Corrections: contact@latenticlabs.com.