# INSTA-YOLO: Real-Time Instance Segmentation

**Authors:** Eslam Mohamed, Abdelrahman Shaker, Hazem Rashed, Ahmad El-Sallab, Mayada Hadhoud  
**arXiv:** [2102.06777](https://arxiv.org/abs/2102.06777)  
**Date:** 2021-02-12  
**Summary page:** https://www.summarizepaper.com/en/paper/2102.06777v1/

## Key points

- Instance segmentation is gaining attention in computer vision applications
- Traditional methods use a two-stage pipeline, but it can be computationally expensive
- The paper proposes a one-stage deep learning model called Insta-YOLO
- Insta-YOLO predicts instances as object contours represented by 2D points
- This approach improves efficiency and real-time performance without sacrificing accuracy
- Experiments on Carvana, Cityscapes, and Airbus datasets show competitive accuracy with faster processing speed on a GTX 1080 GPU
- Insta-YOLO is an efficient and accurate solution for real-time instance segmentation
- Object contours instead of pixel-wise prediction allow for faster processing without compromising accuracy

## Summary

Instance segmentation has gained significant attention in various computer vision applications. It involves assigning different IDs to different objects in a scene, even if they belong to the same class. Traditionally, instance segmentation is performed as a two-stage pipeline, where an object is first detected and then semantic segmentation is applied within the detected box area. However, this approach can be computationally expensive due to the need for costly up-sampling. In this paper titled "INSTA-YOLO: Real-Time Instance Segmentation," the authors propose a novel one-stage end-to-end deep learning model called Insta-YOLO. Unlike traditional methods that rely on pixel-wise prediction, Insta-YOLO predicts instances as object contours represented by 2D points in Cartesian space. This approach offers several advantages such as improved efficiency and real-time performance without sacrificing accuracy. To evaluate the effectiveness of their model, the authors conduct experiments on three datasets: Carvana, Cityscapes and Airbus. They compare their results with state-of-the art models for instance segmentation and demonstrate that Insta YOLO achieves competitive accuracy in terms of mean Average Precision (mAP) while running at twice the speed on a GTX 1080 GPU. Overall, this paper presents a promising solution for real time instance segmentation by introducing Insta YOLO as an efficient and accurate deep learning model. The use of object contours instead of pixel wise prediction allows for faster processing without compromising accuracy. The experimental results validate the effectiveness of Insta YOLO across multiple datasets and highlight its potential for various computer vision applications.