AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

GenRecal: Generation after Recalibration from Large to Small Vision-Language Models

Byung-Kwan Lee, Ryo Hachiuma, Yong Man Ro, Yu-Chiang Frank Wang, Yueh-Hua Wu

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: VLM capability analysis · 10 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
distill with GenRecal from InternVL2.5-78BInternVL2.5-8B · MMMUaccuracy: 56.0 -> 68.1Improved
replace LLaVA-KD with GenRecal under same token typeQwen2-VL-7B · MMMUaccuracy: 58.2 -> 64.2Improved
add first-stage regularization (ablation)GenRecal (InternVL2.5-8B) · MMMUaccuracy: 58.9 -> 68.1Improved
add first-stage regularization (ablation)GenRecal (InternVL2.5-1B) · MMMUaccuracy: 41.1 -> 45.6Improved
add autoregressive loss to KL loss (ablation)GenRecal (InternVL2.5-8B) · Avg over 11 benchmarksaverage accuracy: 59.8 -> 77.1Improved
add new positional embedding (NPE) (ablation)GenRecal (Qwen2-VL-7B) · Avg of MMB/MathVista/MM-Vet/MMMUaverage accuracy: 66.8 -> 73.3Improved
increase Recalibrator depth from 2 to 20 (ablation)GenRecal · Avg of MMB/MathVista/MM-Vet/MMMUaverage accuracy: 76.4 -> 76.7No clear effect
reduce Recalibrator depth from 2 to 1 (ablation)GenRecal · Avg of MMB/MathVista/MM-Vet/MMMUaverage accuracy: 76.4 -> 74.9Hurt
increase training data from 5M to 9M (ablation)GenRecal · Avg of MMB/MathVista/MM-Vet/MMMUaverage accuracy: 76.0 -> 76.4No clear effect
increase training data from 1M to 9M (ablation)GenRecal · Avg of MMB/MathVista/MM-Vet/MMMUaverage accuracy: 70.8 -> 76.4Improved
use GenRecal instead of cross-tokenizer MOT (ablation)InternVL2.5-8B · Avg of MMB/MathVista/MM-Vet/MMMUaverage accuracy: 70.8 -> 76.4Improved
use a stronger teacher VLM (ablation)GenRecal · MMMUaccuracy: 60.3 (NVLM-72B) -> 68.1 (InternVL2.5-78B)Improved

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