Pith. sign in

REVIEW 18 cited by

How Many Unicorns Are in This Image? A Safety Evaluation Benchmark for Vision LLMs

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2311.16101 v1 pith:J56HUPFN submitted 2023-11-27 cs.CV cs.CLcs.LG

Haoqin Tu , Chenhang Cui , Zijun Wang , Yiyang Zhou , Bingchen Zhao , Junlin Han , Wangchunshu Zhou , Huaxiu Yao
show 1 more author
Cihang Xie
This is my paper · ORCID
classification cs.CVcs.CLcs.LG
keywords vllmsevaluationsafetyvisionadversarialllmsperformancerobustness
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

This work focuses on the potential of Vision LLMs (VLLMs) in visual reasoning. Different from prior studies, we shift our focus from evaluating standard performance to introducing a comprehensive safety evaluation suite, covering both out-of-distribution (OOD) generalization and adversarial robustness. For the OOD evaluation, we present two novel VQA datasets, each with one variant, designed to test model performance under challenging conditions. In exploring adversarial robustness, we propose a straightforward attack strategy for misleading VLLMs to produce visual-unrelated responses. Moreover, we assess the efficacy of two jailbreaking strategies, targeting either the vision or language component of VLLMs. Our evaluation of 21 diverse models, ranging from open-source VLLMs to GPT-4V, yields interesting observations: 1) Current VLLMs struggle with OOD texts but not images, unless the visual information is limited; and 2) These VLLMs can be easily misled by deceiving vision encoders only, and their vision-language training often compromise safety protocols. We release this safety evaluation suite at https://github.com/UCSC-VLAA/vllm-safety-benchmark.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 18 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. PRISM: Priority-aware Rubric Internalization via Structured Multimodal Data Synthesis

    cs.LG 2026-08 conditional novelty 7.0 of 10

    PRISM trains MLLMs to act as rubric executors by synthesizing typed, prioritized rules and verification traces, lifting Qwen3-VL-4B from 9.5% to 30.1% Strict accuracy on the authors' PRISM-Eval benchmark.

  2. D-Fusion: Direct Preference Optimization for Aligning Diffusion Models with Visually Consistent Samples

    cs.CV 2025-05 conditional novelty 7.0 of 10

    Mask-guided self-attention fusion creates well-aligned target images that stay visually close to poorly-aligned base images, with full denoising trajectories, and DPO on these pairs improves alignment.

  3. Two Sides of the Same Coin: Co-Evolving Search for Cross-Task Attacks on Vision-Language Models

    cs.CV 2026-08 conditional novelty 6.0 of 10

    CoEvoAttack uses evolutionary search on both text and image sides to generate object-region adversarial examples that transfer across captioning, detection, region categorization, and localization in unified VLMs.

  4. Twins: Learn to Predict Unified Representations with Focal Loss

    cs.CV 2026-07 conditional novelty 6.0 of 10

    Channel-wise concatenation of SigLIP2 and Flux VAE features into one token, trained with a focal-style flow-matching loss, yields a unified representation with 1.59 gFID on ImageNet 256 and VAE-level reconstruction.

  5. One Object, Multiple Lies: A Benchmark for Cross-task Adversarial Attack on Unified Vision-Language Models

    cs.CV 2025-07 conditional novelty 6.0 of 10

    A single adversarial image can make a unified vision-language model misclassify the same object across captioning, detection, region classification, and localization, and the new CrossVLAD benchmark and CRAFT attack m...

  6. MLA-Trust: Benchmarking Trustworthiness of Multimodal LLM Agents in GUI Environments

    cs.AI 2025-06 conditional novelty 6.0 of 10

    MLA-Trust introduces 34 tasks and an evaluation toolbox showing that GUI-interacting multimodal agents are substantially less trustworthy than static multimodal chat models.

  7. Retrieval Visual Contrastive Decoding to Mitigate Object Hallucinations in Large Vision-Language Models

    cs.CV 2025-05 conditional novelty 6.0 of 10

    RVCD uses YOLO detections and retrieved single-concept AI images to adjust LVLM logits at decode time, cutting CHAIR hallucination rates by roughly half versus prior contrastive decoding baselines.

  8. Eagle 2: Building Post-Training Data Strategies from Scratch for Frontier Vision-Language Models

    cs.CV 2025-01 conditional novelty 6.0 of 10

    Eagle2-9B matches or outperforms much larger vision-language models on many benchmarks through a carefully constructed post-training data strategy.

  9. Are VLMs Ready for Autonomous Driving? An Empirical Study from the Reliability, Data, and Metric Perspectives

    cs.CV 2025-01 conditional novelty 6.0 of 10

    A new benchmark with clean, corrupted, and text-only driving inputs shows that vision-language models can answer many driving questions without visual information, so standard accuracy metrics overestimate visual grounding.

  10. Spot Risks Before Speaking! Unraveling Safety Attention Heads in Large Vision-Language Models

    cs.LG 2025-01 conditional novelty 6.0 of 10

    A small set of attention heads in LVLMs flags malicious prompts during the first token; a logistic-regression detector built on them reduces jailbreak success to 1-5%.

  11. ONEBench to Test Them All: Sample-Level Benchmarking Over Open-Ended Capabilities

    cs.LG 2024-12 conditional novelty 6.0 of 10

    ONEBench treats each benchmark sample as a voter in a Plackett-Luce aggregation, enabling open-ended, capability-specific, and incomplete-data model rankings.

  12. Adversarial Attacks against Closed-Source MLLMs via Feature Optimal Alignment

    cs.CV 2025-05 conditional novelty 5.0 of 10

    FOA-Attack aligns global and clustered local features via optimal transport with dynamic ensemble weighting to create targeted adversarial images that transfer to closed-source multimodal LLMs.

  13. Dynamic-VLM: Simple Dynamic Visual Token Compression for VideoLLM

    cs.CV 2024-12 conditional novelty 5.0 of 10

    A video LLM that adaptively compresses each frame's visual tokens, trained on 2M synthetic GPT-generated QA pairs, outperforms LLaVA-OneVision on VideoMME and MuirBench.

  14. Heuristic-Induced Multimodal Risk Distribution Jailbreak Attack for Multimodal Large Language Models

    cs.CR 2024-12 conditional novelty 5.0 of 10

    A black-box jailbreak method distributes harmful semantics across text and image inputs and uses heuristic search to induce multimodal LLMs to answer harmful queries.

  15. Jailbreak Attacks and Defenses against Multimodal Generative Models: A Survey

    cs.CV 2024-11 conditional novelty 5.0 of 10

    A survey that taxonomizes multimodal jailbreak attacks and defenses into four lifecycle levels (input, encoder, generator, output) across Any-to-Text, Any-to-Vision, and Any-to-Any generative models.

  16. A Survey of Safety on Large Vision-Language Models: Attacks, Defenses and Evaluations

    cs.CR 2025-02 conditional novelty 4.0 of 10

    A survey of LVLM safety that adds a lifecycle taxonomy and new benchmark results showing Janus-Pro-7B has weaker safety than several open-source LVLMs.

  17. MME-Survey: A Comprehensive Survey on Evaluation of Multimodal LLMs

    cs.CV 2024-11 conditional novelty 4.0 of 10

    A broad survey that organizes MLLM evaluation benchmarks into capability categories, explains benchmark construction and scoring methods, and identifies gaps in current evaluation practice.

  18. The Scales of Justitia: A Comprehensive Survey on Safety Evaluation of LLMs

    cs.CL 2025-06 conditional novelty 3.0 of 10

    A structured survey of LLM safety evaluation that proposes a why/what/where/how taxonomy and catalogs metrics, datasets, benchmarks, evaluators, and frameworks.

Pith tools