AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

DASH: Detection and Assessment of Systematic Hallucinations of VLMs

Maximilian Augustin, Yannic Neuhaus, Matthias Hein

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Decoding-time hallucination mitigation · 7 citations (OpenAlex)

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 changedWhereMeasuredEffect
use optimized image queries (DASH-OPT) instead of text queries (DASH-LLM) (ablation)DASH on PaliGemma · 380 object classestotal images found: 99.3K -> 221.7KImproved
use optimized image queries (DASH-OPT) instead of text queries (DASH-LLM) (ablation)DASH on PaliGemma · 380 object classestotal clusters: 1892 -> 3895Improved
use optimized image queries (DASH-OPT) instead of text queries (DASH-LLM) (ablation)DASH on LLaVA-NeXT Vicuna · 380 object classestotal clusters: 3632 -> 4632Improved
use optimized image queries (DASH-OPT) instead of text queries (DASH-LLM) (ablation)DASH on LLaVA-NeXT Mistral · 380 object classestotal images found: 78.5K -> 133.8KImproved
use optimized DASH-OPT images instead of DASH-LLM for transfer to other VLMs (ablation)DASH images from PaliGemma · transfer to LLaVA-NeXT Vicunatransfer rate: 0.49 -> 0.43Hurt
swap LLaVA-NeXT LLM backbone from Llama 3.0 to VicunaLLaVA-NeXT · transfer of PaliGemma DASH-LLM imagestransfer rate: 0.26 -> 0.49Hurt
swap Prismatic vision encoder from SigLIP to CLIPPrismatic Vicuna · DASH image transferaverage transfer rate: 0.43 -> 0.60Hurt
scale Qwen2-VL language model from 7B to 72BQwen2-VL · DASH image transferaverage transfer rate: 0.19 -> 0.18No clear effect
fine-tune PaliGemma on DASH imagesPaliGemma · DASH-Baccuracy: 56.4% -> 68.0%Improved
fine-tune PaliGemma on DASH imagesPaliGemma · Amber Existenceaccuracy: 93.2% -> 95.4%Improved
fine-tune PaliGemma on DASH imagesPaliGemma · R-Benchaccuracy: 79.9% -> 80.2%No clear effect
fine-tune PaliGemma on DASH imagesPaliGemma · POPEaccuracy: 87.2% -> 86.4%Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy