Composable Augmentation Encoding for Video Representation Learning

TL;DR

This paper introduces a novel contrastive learning framework for videos that explicitly encodes augmentation parameters, improving the representation's ability to capture temporal and spatial information and enhancing performance on video benchmarks.

Contribution

It proposes Composable Augmentation Encoding (CATE), a method that explicitly incorporates augmentation parameters into contrastive learning for better video representations.

Findings

Achieves state-of-the-art results on multiple video benchmarks.

Encodes valuable spatial and temporal augmentation information.

Improves downstream task performance by capturing invariances.

Abstract

We focus on contrastive methods for self-supervised video representation learning. A common paradigm in contrastive learning is to construct positive pairs by sampling different data views for the same instance, with different data instances as negatives. These methods implicitly assume a set of representational invariances to the view selection mechanism (eg, sampling frames with temporal shifts), which may lead to poor performance on downstream tasks which violate these invariances (fine-grained video action recognition that would benefit from temporal information). To overcome this limitation, we propose an 'augmentation aware' contrastive learning framework, where we explicitly provide a sequence of augmentation parameterisations (such as the values of the time shifts used to create data views) as composable augmentation encodings (CATE) to our model when projecting the video representations for contrastive learning. We show that representations learned by our method encode valuable information about specified spatial or temporal augmentation, and in doing so also achieve state-of-the-art performance on a number of video benchmarks.