PIVOT: Prompting for Video Continual Learning

TL;DR

PIVOT is a novel prompting-based method for video continual learning that leverages pre-trained image models, reducing training complexity and forgetting, and significantly outperforms existing methods on ActivityNet.

Contribution

Introduces PIVOT, the first prompting approach for video continual learning that uses pre-trained image models without in-domain pre-training.

Findings

PIVOT improves state-of-the-art by 27% on ActivityNet.

Effective use of prompting reduces forgetting in continual learning.

Leverages pre-trained models to minimize trainable parameters.

Abstract

Modern machine learning pipelines are limited due to data availability, storage quotas, privacy regulations, and expensive annotation processes. These constraints make it difficult or impossible to train and update large-scale models on such dynamic annotated sets. Continual learning directly approaches this problem, with the ultimate goal of devising methods where a deep neural network effectively learns relevant patterns for new (unseen) classes, without significantly altering its performance on previously learned ones. In this paper, we address the problem of continual learning for video data. We introduce PIVOT, a novel method that leverages extensive knowledge in pre-trained models from the image domain, thereby reducing the number of trainable parameters and the associated forgetting. Unlike previous methods, ours is the first approach that effectively uses prompting mechanisms for continual learning without any in-domain pre-training. Our experiments show that PIVOT improves state-of-the-art methods by a significant 27% on the 20-task ActivityNet setup.