# Segment Everything Everywhere All at Once

**Authors:** Xueyan Zou, Jianwei Yang, Hao Zhang, Feng Li, Linjie Li, Jianfeng Gao, Yong Jae Lee  
**arXiv:** [2304.06718](https://arxiv.org/abs/2304.06718)  
**Date:** 2023-04-13  
**Summary page:** https://www.summarizepaper.com/en/paper/2304.06718v1/

## Key points

- The demand for interactive AI systems is growing rapidly
- Comprehensive studies on human-AI interaction in visual understanding, particularly in segmentation, are needed
- Xueyan Zou and colleagues present SEEM, a promptable and interactive model for segmenting everything everywhere all at once in an image
- SEEM has four key features: versatility, compositionality, interactivity, and semantic-awareness
- The model introduces a versatile prompting engine that supports different types of prompts such as points, boxes, scribbles, masks, texts and referred regions of another image
- SEEM incorporates learnable memory prompts to retain dialog history information via mask-guided cross-attention while using a text encoder to encode text queries and mask labels for open-vocabulary segmentation
- The authors emphasize the potential of their approach to improve human-AI interaction in visual understanding tasks such as segmentation

## Summary

The demand for interactive AI systems is growing rapidly, and there is a need for comprehensive studies on human-AI interaction in visual understanding, particularly in segmentation. To address this gap, Xueyan Zou and colleagues present SEEM, a promptable and interactive model for segmenting everything everywhere all at once in an image. SEEM has four key features: versatility, compositionality, interactivity, and semantic-awareness. The model introduces a versatile prompting engine that supports different types of prompts such as points, boxes, scribbles, masks, texts and referred regions of another image. It also learns a joint visual-semantic space for visual and textual prompts to compose queries on the fly for inference. Additionally, SEEM incorporates learnable memory prompts to retain dialog history information via mask-guided cross-attention while using a text encoder to encode text queries and mask labels for open-vocabulary segmentation. The authors emphasize the potential of their approach to improve human-AI interaction in visual understanding tasks such as segmentation.