Temporal DINO: A Self-supervised Video Strategy to Enhance Action Prediction

TL;DR

Temporal DINO introduces a self-supervised video strategy that leverages teacher-student models to improve action prediction by capturing long-term dependencies, reducing data requirements, and streamlining training.

Contribution

The paper presents a novel self-supervised approach using teacher-student models for action prediction, enhancing performance across multiple architectures with less data and training complexity.

Findings

Achieved an average of 9.9% improvement in prediction accuracy.

Effective across various backbone architectures like 3D-ResNet, Transformer, LSTM.

Reduces reliance on extensive labeled data and complex augmentations.

Abstract

The emerging field of action prediction plays a vital role in various computer vision applications such as autonomous driving, activity analysis and human-computer interaction. Despite significant advancements, accurately predicting future actions remains a challenging problem due to high dimensionality, complex dynamics and uncertainties inherent in video data. Traditional supervised approaches require large amounts of labelled data, which is expensive and time-consuming to obtain. This paper introduces a novel self-supervised video strategy for enhancing action prediction inspired by DINO (self-distillation with no labels). The Temporal-DINO approach employs two models; a 'student' processing past frames; and a 'teacher' processing both past and future frames, enabling a broader temporal context. During training, the teacher guides the student to learn future context by only observing past frames. The strategy is evaluated on ROAD dataset for the action prediction downstream task using 3D-ResNet, Transformer, and LSTM architectures. The experimental results showcase significant improvements in prediction performance across these architectures, with our method achieving an average enhancement of 9.9% Precision Points (PP), highlighting its effectiveness in enhancing the backbones' capabilities of capturing long-term dependencies. Furthermore, our approach demonstrates efficiency regarding the pretraining dataset size and the number of epochs required. This method overcomes limitations present in other approaches, including considering various backbone architectures, addressing multiple prediction horizons, reducing reliance on hand-crafted augmentations, and streamlining the pretraining process into a single stage. These findings highlight the potential of our approach in diverse video-based tasks such as activity recognition, motion planning, and scene understanding.