ASCNet: Self-supervised Video Representation Learning with Appearance-Speed Consistency - Invited скачать в хорошем качестве

ASCNet: Self-supervised Video Representation Learning with Appearance-Speed Consistency - Invited

Self-supervised video representation learning aims to learn video features from unlabeled videos. The learned model weights can be transferred to downstream tasks, such as action recognition and video retrieval. In this paper, we observe that the consistency between positive samples is the key to learn robust video representation. Specifically, we propose two tasks to learn appearance and speed consistency, respectively. The appearance consistency perception (ACP) task aims to maximize the similarity between two clips of the same video with different playback speeds. The speed consistency perception (SCP) task aims to maximize the similarity between two clips with the same playback speed but different appearance information. We show that optimizing the two tasks jointly consistently improves the performance on downstream tasks.