InternGPT: Solving Vision-Centric Tasks by Interacting with ChatGPT Beyond Language

Zhaoyang Liu^?1, Yinan He^?1, Wenhai Wang^??1, Weiyun Wang^?1, Yi Wang^?1, Shoufa Chen^?2,1 Qinglong Zhang^?1, Yang Yang¹, Qingyun Li¹, Jiashuo Yu¹, Kunchang Li^3,1, Zhe Chen^4,1, Xue Yang¹, Xizhou Zhu^5,1, Yali Wang^3,1, Limin Wang^4,1, Ping Luo^2,1, Jifeng Dai^6,1, Yu Qiao¹

 ¹OpenGVLab, Shanghai AI Laboratory     ²The University of Hong Kong     ³Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences     ⁴Nanjing University     ⁵SenseTime Research ⁶Tsinghua University

https://github.com/OpenGVLab/InternGPT 

We’re going to use the best pointing device in the world. We’re going to use a pointing device that we’re all born with — born with ten of them. We’re going to use our fingers. We’re going to touch this with our fingers. — Steve Jobs

Abstract

We present an interactive visual framework named InternGPT, or iGPT for short. The framework integrates chatbots that have planning and reasoning capabilities, such as ChatGPT, with non-verbal instructions like pointing movements that enable users to directly manipulate images or videos on the screen. Pointing (including gestures, cursors, etc.) movements can provide more flexibility and precision in performing vision-centric tasks that require fine-grained control, editing, and generation of visual content. The name InternGPT stands for interaction, nonverbal, and chatbots. Different from existing interactive systems that rely on pure language, by incorporating pointing instructions, the proposed iGPT significantly improves the efficiency of communication between users and chatbots, as well as the accuracy of chatbots in vision-centric tasks, especially in complicated visual scenarios where the number of objects is greater than 2. Additionally, in iGPT, an auxiliary control mechanism is used to improve the control capability of LLM, and a large vision-language model termed Husky is fine-tuned for high-quality multi-modal dialogue (impressing ChatGPT-3.5-turbo with 93.89% GPT-4 Quality). We hope this work can spark new ideas and directions for future interactive visual systems.