In this paper, we propose Recycling+, an algorithm that uses temporal information for video frame classification with noisy label. With the proposed algorithm, the network can be trained robustly through not only clean samples with original labels but also noisy samples with corrected labels obtained by temporal average predictions.

中文翻译：

---

带噪声标签的学习算法用于视频帧分类

在本文中，我们提出了Recycling +算法，该算法将时间信息用于带有噪声标签的视频帧分类。使用提出的算法，不仅可以通过带有原始标签的干净样本，而且可以通过通过时间平均预测获得的带有经过校正的标签的噪声样本，对网络进行鲁棒的训练。