Benchmarking Unsupervised Object Representations for Video Sequences

该论文设计了一个包含四个不同复杂度数据集和七个额外测试集的基准，用于评估无监督对象表示方法在检测、前景背景分割和跟踪等基本感知能力上的表现。

This paper develops a benchmark consisting of four datasets with distinct complexity levels and seven additional test sets, which is designed to evaluate the performance of unsupervised object representation methods on fundamental perceptual capabilities including object detection, foreground-background segmentation, and visual tracking.