视频中的自监督学习：预测运动与外观统计

Self-supervised learning of spatio-temporal representation in videos through the prediction of motion patterns and appearance statistics.

2019年发表于CVPR的一篇论文中,研究者开发了一个预设任务,该任务通过从运动学特征和色彩学特征提取出关键统计信息,并结合最大动作位置及方向等关键参数进行分析.具体而言,该预设任务涉及从视频数据中提取出物体运动幅度最大的区域及其方向信息,同时结合色彩变化的最大与最小区域及其色调数值等关键参数进行建模.在论文引言部分,作者指出视觉系统中动作表征主要依赖于预训练识别模型.

The research draws inspiration from Giese and Poggio's work on human visual systems [14], where it is revealed that motion representation is based on a collection of learned patterns.

Pattern representations are established through a series of static body shape captures f