VCAS-Motion (Video Class Agnostic Segmentation Benchmark)

视频类不可知分割（VCAS）是分割对象的任务，不考虑其语义，结合单目视频序列的外观、运动和几何。这背后的主要动机是考虑场景中的未知对象，并作为冗余信号以及已知类的分割以提高安全性，如下图所示。此 VCAS 基准测试由 KITTI-MOTS 和 Cityscapes-VPS 构建。

Video Class-Agnostic Segmentation (VCAS) refers to the task of segmenting objects without considering their semantics, which integrates appearance, motion and geometry information from monocular video sequences. The core motivation behind this task is to account for unknown objects in the scene, serve as redundant signals alongside the segmentation of known classes to enhance safety, as shown in the following figure. This VCAS benchmark is constructed based on KITTI-MOTS and Cityscapes-VPS.

搜集汇总

背景与挑战

背景概述

VCAS-Motion是一个视频类不可知分割基准数据集，专注于自动驾驶场景中不考虑语义的对象分割任务，利用外观、运动和几何信息。该数据集基于KITTI-MOTS和Cityscapes-VPS构建，于2021年发布，旨在提升未知对象分割的安全性和冗余信号处理。