Recurrent Convolutions for Causal 3D CNNs
This addresses the need for causal video processing in online settings, offering a novel architecture that overcomes limitations of existing 3D CNNs.
The paper tackles the problem of anti-causal 3D CNNs in video processing by proposing a Recurrent Convolutional Network (RCN) that decomposes 3D convolutions into spatial and temporal components, enabling causal outputs, flexible temporal reasoning, and temporal resolution preservation. Experiments on Kinetics and MultiThumos datasets show it performs comparably to anti-causal methods while using fewer parameters.
Recently, three dimensional (3D) convolutional neural networks (CNNs) have emerged as dominant methods to capture spatiotemporal representations in videos, by adding to pre-existing 2D CNNs a third, temporal dimension. Such 3D CNNs, however, are anti-causal (i.e., they exploit information from both the past and the future frames to produce feature representations, thus preventing their use in online settings), constrain the temporal reasoning horizon to the size of the temporal convolution kernel, and are not temporal resolution-preserving for video sequence-to-sequence modelling, as, for instance, in action detection. To address these serious limitations, here we present a new 3D CNN architecture for the causal/online processing of videos. Namely, we propose a novel Recurrent Convolutional Network (RCN), which relies on recurrence to capture the temporal context across frames at each network level. Our network decomposes 3D convolutions into (1) a 2D spatial convolution component, and (2) an additional hidden state $1\times 1$ convolution, applied across time. The hidden state at any time $t$ is assumed to depend on the hidden state at $t-1$ and on the current output of the spatial convolution component. As a result, the proposed network: (i) produces causal outputs, (ii) provides flexible temporal reasoning, (iii) preserves temporal resolution. Our experiments on the large-scale large Kinetics and MultiThumos datasets show that the proposed method performs comparably to anti-causal 3D CNNs, while being causal and using fewer parameters.