möglich sobald bei der ZB eingereicht worden ist.
Training-Free Uncertainty Guidance for Complex Visual Tasks with MLLMs.
In: (Computer Vision – ECCV 2026). Berlin [u.a.]: Springer, 2026. 606-624 (Lect. Notes Comput. Sc. ; 17006 LNCS)
Multimodal Large Language Models (MLLMs) often struggle with fine-grained perception, such as identifying small objects in high-resolution images or detecting key moments in long videos. Existing methods typically rely on complex, task-specific fine-tuning, which reduces generalizability and increases system complexity. In this work, we propose an effective, training-free framework that uses an MLLM’s intrinsic uncertainty as proactive guidance. Our core insight is that a model’s uncertainty decreases when provided with relevant visual information. We introduce a unified mechanism that scores candidate visual inputs by response uncertainty, enabling the model to autonomously focus on the most informative data. We apply this simple principle to three challenging visual tasks: Visual Search, Long Video Understanding, and Temporal Grounding, allowing off-the-shelf MLLMs to achieve performance competitive with specialized, fine-tuned systems. Our results demonstrate that leveraging intrinsic uncertainty is a powerful strategy for improving fine-grained multimodal performance. Code is available at https://github.com/ExplainableML/ug-framework.
Altmetric
Weitere Metriken?
Zusatzinfos bearbeiten
[➜Einloggen]
Publikationstyp
Artikel: Konferenzbeitrag
Schlagwörter
Fine-grained Perception ; Multimodal Large Language Models ; Uncertainty Estimation
ISSN (print) / ISBN
0302-9743
e-ISSN
1611-3349
Konferenztitel
Computer Vision – ECCV 2026
Zeitschrift
Lecture Notes in Computer Science
Quellenangaben
Band: 17006 LNCS,
Seiten: 606-624
Verlag
Springer
Verlagsort
Berlin [u.a.]