Figureure 10 · Qualitative unsupervised VPS examples on DAVIS [132]Figure 10. Qualitative unsupervised VPS examples on DAVIS [132]. We provide qualitative samples for VideoCUPS inference on DAVIS videos using the Cityscapes-VPS class assignments for visualization purposes. VideoCUPS generalizes to the unseen dataset and even to unseen semantic concepts.这张可视化用来解释 Scene-Centric Unsupervised Video Panoptic Segmentation 学到的中间表征或对齐关系。重点看它是否支持正文里的机制判断。Figureure 7 · KITTI-STEP—Qualitative unsupervised VPS examplesFigure 7. KITTI-STEP—Qualitative unsupervised VPS examples. We compare our proposed method VideoCUPS to the proposed baselines DepthG [91] + VideoCutLER [110], U2Seg [81] + SORT, CUPS [41] + SORT [9], and CUPS† [41] + SORT [9] on KITTI-STEP [114] val.这张图来自论文 PDF 的结构化抽取。当前用于辅助理解 Scene-Centric Unsupervised Video Panoptic Segmentation 的方法或实验,请结合正文精读段落一起看。