Research map / Vision-language and multimodal / Multimodal LLMs
Do Vision Language Models Recognize Visual Ambiguity?
Huy Ta, Trang Nguyen, Townim Chowdhury, ANKIT YADAV, Minh-Son To, Zhibin Liao, Johan Verjans, Hieu Phan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add image-only perturbation to Semantic Entropy (ablation) | SE · AOKVQA (Qwen2.5-VL) | AUC: 0.702 -> 0.761 | Improved |
| Add image-only perturbation to SNNE (ablation) | SNNE · AOKVQA (Gemma3) | AUC: 0.651 -> 0.746 | Improved |
| Use prototype semantic aggregation instead of SE entropy, with visual perturbation (ablation) | VSE · AOKVQA (Qwen2.5-VL) | AUC: 0.761 -> 0.783 | Improved |
| Use prototype aggregation without visual perturbation (ablation) | ProtoSem · AOKVQA (Qwen2.5-VL) | AUC: 0.711 -> 0.783 | Improved |
| Perturb image only rather than text and image (VL-Uncertainty) | VSE · AOKVQA (Qwen2.5-VL) | AUC: 0.756 -> 0.783 | Improved |
| Use VSE instead of SE on adversarial VILP | VSE · VILP (Qwen2.5-VL) | AUC: 0.535 -> 0.650 | Improved |
| Use VSE instead of best baseline on VLM-are-biased | VSE · VLM-are-biased (Qwen2.5-VL) | AUC: 0.783 -> 0.826 | Improved |
| Use VSE instead of SE on MMVet | VSE · MMVet (Qwen2.5-VL) | AUC: 0.716 -> 0.781 | Improved |
| Use VSE instead of SE on OKVQA | VSE · OKVQA (Qwen2.5-VL) | AUC: 0.758 -> 0.767 | Improved |
| Text paraphrasing vs image perturbation as source of output variability | Joint perturbation · VILP (Qwen2.5-VL) | cluster purity: PT 54.3 vs PI 38.8 | Mixed |
| Stochastic decoding alone on visually confident samples | Semantic Entropy · VILP (Qwen2.5-VL) | semantic entropy | Hurt |
| Use VSE vs strongest baseline on Gemma3 VILP | VSE · VILP (Gemma3) | AUC: 0.660 -> 0.665 | No clear effect |
Closest papers by meaning
- Synthesize Diagnose and Optimize: Towards Fine-Grained Vision-Language Understanding (CVPR 2024 · similarity 0.89)
- Revisiting the Role of Language Priors in Vision-Language Models (ICML 2024 · similarity 0.89)
- Aligning What Vision-Language Models See and Perceive with Adaptive Information Flow (CVPR 2026 · similarity 0.89)
- Enhancing Vision-Language Model Reliability with Uncertainty-Guided Dropout Decoding (NeurIPS 2025 · similarity 0.89)
- When to Think and When to Look: Uncertainty-Guided Lookback (CVPR 2026 · similarity 0.89)
- Exploiting the Asymmetric Uncertainty Structure of Pre-trained VLMs on the Unit Hypersphere (NeurIPS 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
