Systematic Literature Review of Modern Instance segmentation : Dari CNN hingga Transformer dan Open-Vocabulary Models

Abstract

This study examines the development of instance segmentation  methods in the field of Computer vision through a Systematic Literature Review approach. The main issue in this study is the rapid growth of instance segmentation  methods, which has led to a wide variety of models, techniques, and application domains, necessitating a systematic analysis to understand trends and the direction of their development. The research was conducted using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines through a process of identifying, selecting, and analyzing 30 articles fully indexed in SCOPUS and sourced from various academic databases such as ScienceDirect, SpringerLink, IEEE Xplore, arXiv, MDPI, Wiley Online Library, ACM Digital Library, and Nature Scientific Reports. The study reveals that deep learning-based techniques such as Convolutional Neural Networks, YOLO, Transformers, diffusion models, and vision-language models have advanced significantly in improving segmentation accuracy, computational efficiency, and model generalization capabilities. Additionally, the use of synthetic data, multimodal learning, and open-vocabulary techniques has emerged as a major trend in modern instance segmentation  development. This research provides an in-depth explanation of technological evolution, research challenges, and opportunities for the future development of instance segmentation  methods.