Real-time human-centric segmentation for complex video scenes
作者：
Highlights：
• A novel one-stage human-centric segmentation framework in complex video scenes.
• Segmenting and tracking all humans states in a complex video.
• The center of the bounding box will deteriorate in heavily occluded conditions.
• An Inner Center Sampling to improve the accuracy of instance segmentation.
• A new benchmark called HVIS, which comprises 1447 human instance masks.
摘要
•A novel one-stage human-centric segmentation framework in complex video scenes.•Segmenting and tracking all humans states in a complex video.•The center of the bounding box will deteriorate in heavily occluded conditions.•An Inner Center Sampling to improve the accuracy of instance segmentation.•A new benchmark called HVIS, which comprises 1447 human instance masks.
论文关键词：Multiple human tracking,Video instance segmentation,One-stage detector,Video understanding,Deep neural networks
论文评审过程：Received 23 October 2021, Revised 6 April 2022, Accepted 28 August 2022, Available online 8 September 2022, Version of Record 15 September 2022.
论文官网地址：https://doi.org/10.1016/j.imavis.2022.104552
