AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Training-free Uncertainty Guidance for Complex Visual Tasks with MLLMs

Sanghwan Kim, Rui Xiao, Stephan Alaniz, Yongqin Xian, Zeynep Akata

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Instruction-tuned multimodal assistants

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
add UG-Search (lowest-entropy crop selection) to the MLLMInternVL2.5-8B · V* Benchaccuracy: 71.7 -> 91.1Improved
add UG-Search to standard QA with regular-size imagesQwen2.5-VL-7B · TextVQAaccuracy: 76.1 -> 79.7Improved
add UG-Sample (entropy-based top-8 frame selection) instead of uniform samplingQwen2.5-VL-7B · Video-MME (overall)accuracy: 53.7 -> 59.8Improved
add UG-Sample on short-video benchmarksQwen2.5-VL-7B · EgoSchemaaccuracy: 53.7 -> 56.6Improved
add UG-Ground (BRC score with Kadane search) to the MLLMLLaVA-OV-7B · Charades-STAmIoU: 10.4 -> 46.8Improved
replace AKS CLIP scoring with entropy score (UG-Sample+AKS hybrid) for coverageUG-Sample · MLVUaccuracy: 62.0 -> 65.0Improved
use a larger scoring MLLM (26B instead of 1B) with an 8B answerer (ablation)UG-Search · V* Benchaccuracy: 75.4 -> 92.0Improved
use a larger scoring MLLM (26B instead of 1B) with an 8B answerer (ablation)UG-Sample · Video-MMEaccuracy: 58.5 -> 61.8Improved
add YOLOv8 pre-filter to UG-Search (ablation)UG-Search on LLaVA-OV-7B · V* Benchaccuracy: 81.2 -> 74.9Hurt
add SAM2 pre-filter to UG-Search (ablation)UG-Search on LLaVA-OV-7B · V* Benchaccuracy: 81.2 -> 78.2Hurt
add SigLIP pre-filter to UG-Sample (top-32 frames) (ablation)UG-Sample on LLaVA-OV-7B · Video-MMEaccuracy: 58.6 -> 57.3Mixed
add DINOv2 pre-filter to UG-Sample (top-32 frames) (ablation)UG-Sample on LLaVA-OV-7B · Video-MMEaccuracy: 58.6 -> 56.4Mixed

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