Mini-Gemini: Mining the Potential of Multi-modality Vision Language Models
arxiv(2024)
摘要
In this work, we introduce Mini-Gemini, a simple and effective framework
enhancing multi-modality Vision Language Models (VLMs). Despite the
advancements in VLMs facilitating basic visual dialog and reasoning, a
performance gap persists compared to advanced models like GPT-4 and Gemini. We
try to narrow the gap by mining the potential of VLMs for better performance
and any-to-any workflow from three aspects, i.e., high-resolution visual
tokens, high-quality data, and VLM-guided generation. To enhance visual tokens,
we propose to utilize an additional visual encoder for high-resolution
refinement without increasing the visual token count. We further construct a
high-quality dataset that promotes precise image comprehension and
reasoning-based generation, expanding the operational scope of current VLMs. In
general, Mini-Gemini further mines the potential of VLMs and empowers current
frameworks with image understanding, reasoning, and generation simultaneously.
Mini-Gemini supports a series of dense and MoE Large Language Models (LLMs)
from 2B to 34B. It is demonstrated to achieve leading performance in several
zero-shot benchmarks and even surpasses the developed private models. Code and
models are available at https://github.com/dvlab-research/MiniGemini.
更多查看译文
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要