The PanoVOS project aims to address the challenges of video segmentation in panoramic videos. The team at Fudan University and the University of Michigan have created a new dataset, PanoVOS, which includes 150 videos with high-resolution and diverse motions. They also introduced a Panoramic Space Consistency Transformer (PSCFormer) which effectively utilizes semantic boundary information from previous frames for pixel-level matching with the current frame. The PanoVOS dataset and the PSCFormer network offer new opportunities and challenges in panoramic video object segmentation.
Publication date: 21 Sep 2023
Project Page: https://github.com/shilinyan99/PanoVOS
Paper: https://arxiv.org/pdf/2309.12303
