Set-of-mark Prompting Unleashes Extraordinary Visual Grounding In GPT-4V

Jianwei Yang et al.. Arxiv 2023 – 16 citations

[Paper] [Code]    
GPT Reinforcement Learning Prompting Has Code Multimodal Models Model Architecture

We present Set-of-Mark (SoM), a new visual prompting method, to unleash the visual grounding abilities of large multimodal models (LMMs), such as GPT-4V. As illustrated in Fig. 1 (right), we employ off-the-shelf interactive segmentation models, such as SEEM/SAM, to partition an image into regions at different levels of granularity, and overlay these regions with a set of marks e.g., alphanumerics, masks, boxes. Using the marked image as input, GPT-4V can answer the questions that require visual grounding. We perform a comprehensive empirical study to validate the effectiveness of SoM on a wide range of fine-grained vision and multimodal tasks. For example, our experiments show that GPT-4V with SoM in zero-shot setting outperforms the state-of-the-art fully-finetuned referring expression comprehension and segmentation model on RefCOCOg. Code for SoM prompting is made public at: https://github.com/microsoft/SoM.

Similar Work