CAMVOS: Leveraging Context and Memory for Advanced Video Object Segmentation
Year: 2024
Type: conference-paper
Abstract: We introduce CAMVOS (Context-Aware Memory Video Object Segmentation), a novel method designed to enhance video object segmentation by leveraging contextual information, long-range dependency and memory-efficient techniques. CAMVOS utilizes a sophisticated framework that integrates spatio-temporal context from both objects' immediate and extended surroundings, improving the accuracy and robustness of segmentation across varying scenes and motion patterns. Specifically, it extracts the multi-scale features leveraging atrous convolutions and then encodes the contextual relationship between features using the graph attention networks. It also maintains high performance while minimizing computational overhead, making it suitable for real-time applications by incorporating memory-efficient algorithms. Our experiments on three benchmark datasets (YouTube VOS 2018, DAVIS 2016, and DAVIS 2017) demonstrate that CAMVOS significantly outperforms existing state-of-the-art methods in video object segmentation, setting a new benchmark for precision and efficiency.
Authors: Waqas Arshid, Mohammad Awrangjeb, Alan Wee Chung Liew, Yongsheng Gao
Institution: Griffith University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4407474752