Is 3D Convolution with 5D Tensors Really Necessary for Video Analysis?

TL;DR

This paper investigates whether 3D convolutions with 5D tensors are necessary for video analysis, proposing efficient alternatives using 4D and 3D tensors that improve speed and accuracy.

Contribution

The authors introduce novel techniques for implementing 3D convolutional blocks with lower-dimensional tensors, reducing computational cost and supporting edge device deployment.

Findings

Significant speed improvements over traditional 3D convolutions

Enhanced accuracy with fewer parameters

Effective spatio-temporal processing using 4D tensors

Abstract

In this paper, we present a comprehensive study and propose several novel techniques for implementing 3D convolutional blocks using 2D and/or 1D convolutions with only 4D and/or 3D tensors. Our motivation is that 3D convolutions with 5D tensors are computationally very expensive and they may not be supported by some of the edge devices used in real-time applications such as robots. The existing approaches mitigate this by splitting the 3D kernels into spatial and temporal domains, but they still use 3D convolutions with 5D tensors in their implementations. We resolve this issue by introducing some appropriate 4D/3D tensor reshaping as well as new combination techniques for spatial and temporal splits. The proposed implementation methods show significant improvement both in terms of efficiency and accuracy. The experimental results confirm that the proposed spatio-temporal processing structure outperforms the original model in terms of speed and accuracy using only 4D tensors with fewer parameters.