Learning Spatio-Temporal Features with 3D Residual Networks for Action Recognition

TL;DR

This paper introduces 3D Residual Networks (ResNets) for action recognition in videos, demonstrating improved performance and reduced overfitting compared to shallower 3D CNNs, validated on ActivityNet and Kinetics datasets.

Contribution

The paper proposes a novel 3D ResNet architecture for video action recognition, enabling deeper networks that outperform shallow models like C3D without overfitting.

Findings

3D ResNets outperform shallow 3D CNNs like C3D.

Training on Kinetics reduces overfitting despite large model size.

Achieved state-of-the-art results on ActivityNet and Kinetics datasets.

Abstract

Convolutional neural networks with spatio-temporal 3D kernels (3D CNNs) have an ability to directly extract spatio-temporal features from videos for action recognition. Although the 3D kernels tend to overfit because of a large number of their parameters, the 3D CNNs are greatly improved by using recent huge video databases. However, the architecture of 3D CNNs is relatively shallow against to the success of very deep neural networks in 2D-based CNNs, such as residual networks (ResNets). In this paper, we propose a 3D CNNs based on ResNets toward a better action representation. We describe the training procedure of our 3D ResNets in details. We experimentally evaluate the 3D ResNets on the ActivityNet and Kinetics datasets. The 3D ResNets trained on the Kinetics did not suffer from overfitting despite the large number of parameters of the model, and achieved better performance than relatively shallow networks, such as C3D. Our code and pretrained models (e.g. Kinetics and ActivityNet) are publicly available at https://github.com/kenshohara/3D-ResNets.