AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

TWIST & SCOUT: Grounding Multimodal LLM-Experts by Forget-Free Tuning

Aritra Bhowmik, Mohammad Mahdi Derakhshani, Dennis Koelma, Yuki M. Asano, Martin R. Oswald, Cees G. M. Snoek

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Chain-of-thought multimodal reasoning · 3 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
add TWIST twin-expert module (ablation)TWIST & SCOUT · MM-Vetscore: 0 -> 34.3Improved
add TWIST twin-expert module (ablation)TWIST & SCOUT · RefCOCOaccuracy: 84.8 -> 85.6Improved
add TWIST twin-expert module (ablation)TWIST & SCOUT · COCO grounded captioningAP: 9.6 -> 11.4Improved
add SCOUT dataset (ablation)TWIST & SCOUT · RefCOCOaccuracy: 85.6 -> 86.9Improved
add SCOUT dataset (ablation)TWIST & SCOUT · COCO grounded captioningAP: 11.4 -> 13.7Improved
add step-wise loss (ablation)TWIST & SCOUT · COCO grounded captioningAP: 13.7 -> 15.0Improved
add learned alpha gating (alpha=0.31 instead of 0) (ablation)TWIST & SCOUT · COCO grounded captioningmAP: 0 -> 15Improved
use 2 grounding experts instead of 1 (ablation)TWIST & SCOUT · RefCOCO test-Aaccuracy: 79.8 -> 90.2Improved
use 4 grounding experts instead of 2 (ablation)TWIST & SCOUT · RefCOCO test-Aaccuracy: 90.2 -> 90.3No clear effect
use MoE-LLaVA backbone instead of LLaVA (ablation)TWIST & SCOUT · RefCOCO test-Aaccuracy: 85.7 -> 90.2Improved
replace Visual Genome with SCOUT as fine-tuning data (ablation)TWIST & SCOUT · COCO grounded captioningAP50: 38.5 -> 49.3Improved
add negative samples to SCOUT (ablation)TWIST & SCOUT · COCO grounded captioningAP50: 48.1 -> 49.3Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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