PGT: A Progressive Method for Training Models on Long Videos

TL;DR

The paper introduces PGT, a progressive training method that enables end-to-end training of long videos by propagating information sequentially, overcoming computational limitations of traditional clip-based approaches.

Contribution

It proposes a novel progressive training approach inspired by NLP techniques, allowing effective end-to-end training of long videos with limited resources.

Findings

Improves SlowOnly network by 3.7 mAP on Charades

Increases top-1 accuracy by 1.9 on Kinetics

Achieves significant performance gains with negligible overhead

Abstract

Convolutional video models have an order of magnitude larger computational complexity than their counterpart image-level models. Constrained by computational resources, there is no model or training method that can train long video sequences end-to-end. Currently, the main-stream method is to split a raw video into clips, leading to incomplete fragmentary temporal information flow. Inspired by natural language processing techniques dealing with long sentences, we propose to treat videos as serial fragments satisfying Markov property, and train it as a whole by progressively propagating information through the temporal dimension in multiple steps. This progressive training (PGT) method is able to train long videos end-to-end with limited resources and ensures the effective transmission of information. As a general and robust training method, we empirically demonstrate that it yields significant performance improvements on different models and datasets. As an illustrative example, the proposed method improves SlowOnly network by 3.7 mAP on Charades and 1.9 top-1 accuracy on Kinetics with negligible parameter and computation overhead. Code is available at https://github.com/BoPang1996/PGT.