The paper introduces Mini-Gemini, a framework aimed at enhancing Vision Language Models (VLMs) to close the performance gap with advanced models like GPT-4 and Gemini. It focuses on improving visual tokens resolution, creating high-quality datasets for better image comprehension, and expanding VLMs' operational scope. Mini-Gemini supports a range of large language models and has shown superior performance in zero-shot benchmarks. The code and models are publicly available.
Comments
The paper introduces Mini-Gemini, a framework aimed at enhancing Vision Language Models (VLMs) to close the performance gap with advanced models like GPT-4 and Gemini. It focuses on improving visual tokens resolution, creating high-quality datasets for better image comprehension, and expanding VLMs' operational scope. Mini-Gemini supports a range of large language models and has shown superior performance in zero-shot benchmarks. The code and models are publicly available.