AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Mini-Gemini: Mining the Potential of Multi-Modality Vision Language Models

Yanwei Li, Yuechen Zhang, Chengyao Wang, Zhisheng Zhong, Yixin Chen, Ruihang Chu, Shaoteng Liu, Jiaya Jia

TPAMI 2025 · Vision-language and multimodal › Multimodal LLMs · approach: VLM capability analysis · 446 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 patch info mining with ConvNeXt-L HR encoder (LR 224, HR 512) (ablation)LLaVA-1.5 Vicuna-7B · TextVQAaccuracy: 54.1 -> 58.1Improved
add patch info mining (LR 336, HR 768) (ablation)LLaVA-1.5 Vicuna-7B · TextVQAaccuracy: 58.2 -> 61.5Improved
add patch info mining (LR 336, HR 768) (ablation)LLaVA-1.5 Vicuna-7B · MM-Vetscore: 31.1 -> 34.6Improved
raise HR resolution from 512 to 768 (LR 224) (ablation)Mini-Gemini · MMEscore: 1485.2 -> 1478.3Hurt
use ConvNeXt-B instead of ConvNeXt-L as HR encoder (ablation)Mini-Gemini · MMEscore: 1517.0 -> 1451.7Hurt
use ConvNeXt-XXL instead of ConvNeXt-L as HR encoder (ablation)Mini-Gemini · TextVQAaccuracy: 61.5 -> 62.0No clear effect
use ConvNeXt-XXL instead of ConvNeXt-L as HR encoder (ablation)Mini-Gemini · MM-Vetscore: 34.6 -> 33.8Hurt
add ShareGPT4V captions (ablation)Mini-Gemini · TextVQAaccuracy: 61.5 -> 63.2Improved
remove TextCaps data (ablation)Mini-Gemini · TextVQAaccuracy: 63.2 -> 59.0Hurt
add OCR-related data (ablation)Mini-Gemini · TextVQAaccuracy: 58.7 -> 61.6Improved
add generation-related instructions (ablation)Mini-Gemini · MM-Vetscore: 33.7 -> 37.0Improved
add ALLaVA GPT4V responses (ablation)Mini-Gemini · MM-Vetscore: 37.0 -> 40.8Improved

Later papers that compared with this method

3 later accepted papers report Mini-Gemini as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.

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