AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Vary: Scaling up the Vision Vocabulary for Large Vision-Language Models

Haoran Wei, Lingyu Kong, Jinyue Chen, Liang Zhao, Zheng Ge, Jinrong Yang, Jianjian Sun, Chunrui Han, Xiangyu Zhang

ECCV 2024 · Vision-language and multimodal › Multimodal LLMs · approach: Instruction-tuned multimodal assistants · 85 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 new SAM-based vision vocabulary alongside CLIP (Vary-base with Vicuna-7B and LLaVA-665k SFT)LLaVA-1.5-style LVLM · MM-Vettotal score: 30.5 -> 32.9Improved
use Qwen-7B and LLaVA-80k instead of Vicuna-7B and LLaVA-665kVary-base · MM-Vettotal score: 32.9 -> 36.2Improved
scale up vision vocabulary, Vary-base versus Nougat on markdown conversionVary-base · Markdown conversion average (own test set)edit distance: 0.245 -> 0.181Improved
scale up vision vocabulary, Vary-base versus Nougat on English plain-text OCR F1Vary-base · Document OCR English (own test set)F1-score: 89.91 -> 88.24No clear effect
use 665k instead of 80k SFT data (ablation)Vary-base · DocVQA valANLS: 78.2 -> 78.1No clear effect
use 665k instead of 80k SFT data (ablation)Vary-base · ChartQA averagerelaxed accuracy: 65.3 -> 66.1Improved

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