研究概況
對視頻中感興趣的目標進行分割,給定第一幀mask的VOS稱爲one-shot VOS (單樣本VOS),無第一幀目標mask的稱爲zero-shot VOS(無樣本VOS)。視頻目標分割(Video Object Segmentation,VOS) 是集檢測、跟蹤、分割、ReID於一體的計算機視覺任務,提供了更加豐富的信息,標註成本很高,計算量也比較大,近年來隨着高性能設備和相關數據集的出現,也越來越受到關注。CVPR 2020 總計有8篇相關文獻。
論文列表(8篇)
A Transductive Approach for Video Object Segmentation
Code:https://github.com/microsoft/transductive-vos.pytorch
Learning Fast and Robust Target Models for Video Object Segmentation
Code :https://github.com/andr345/frtm-vos
Fast Video Object Segmentation With Temporal Aggregation Network and Dynamic Template Matching
Code: https://xuhuaking.github.io/Fast-VOS-DTTM-TAN/
Learning Video Object Segmentation From Unlabeled Videos
Code:https://github.com/carrierlxk/MuG
State-Aware Tracker for Real-Time Video Object Segmentation
Code:https://github.com/XavierCHEN34/State-Aware-Tracker
解讀:漫談視頻目標跟蹤與分割
Memory Aggregation Networks for Efficient Interactive Video Object Segmentation
Visual-Textual Capsule Routing for Text-Based Video Segmentation
Fast Template Matching and Update for Video Object Tracking and Segmentation
Code: https://github.com/insomnia94/FTMU