TS-LSTM and Temporal-Inception: Exploiting Spatiotemporal Dynamics for Activity Recognition

TL;DR

This paper explores and compares different deep neural network architectures, including RNNs and Temporal-ConvNets, for capturing spatiotemporal dynamics in video-based activity recognition, achieving state-of-the-art results.

Contribution

It systematically evaluates two-stream ConvNets with RNNs and Temporal-ConvNets, proposing two new architectures to better exploit spatiotemporal information in videos.

Findings

Both RNNs and Temporal-ConvNets improve activity recognition performance.

Proper data segmentation is crucial for LSTM-based methods.

Achieved state-of-the-art accuracy on UCF101 and HMDB51 datasets.

Abstract

Recent two-stream deep Convolutional Neural Networks (ConvNets) have made significant progress in recognizing human actions in videos. Despite their success, methods extending the basic two-stream ConvNet have not systematically explored possible network architectures to further exploit spatiotemporal dynamics within video sequences. Further, such networks often use different baseline two-stream networks. Therefore, the differences and the distinguishing factors between various methods using Recurrent Neural Networks (RNN) or convolutional networks on temporally-constructed feature vectors (Temporal-ConvNet) are unclear. In this work, we first demonstrate a strong baseline two-stream ConvNet using ResNet-101. We use this baseline to thoroughly examine the use of both RNNs and Temporal-ConvNets for extracting spatiotemporal information. Building upon our experimental results, we then propose and investigate two different networks to further integrate spatiotemporal information: 1) temporal segment RNN and 2) Inception-style Temporal-ConvNet. We demonstrate that using both RNNs (using LSTMs) and Temporal-ConvNets on spatiotemporal feature matrices are able to exploit spatiotemporal dynamics to improve the overall performance. However, each of these methods require proper care to achieve state-of-the-art performance; for example, LSTMs require pre-segmented data or else they cannot fully exploit temporal information. Our analysis identifies specific limitations for each method that could form the basis of future work. Our experimental results on UCF101 and HMDB51 datasets achieve state-of-the-art performances, 94.1% and 69.0%, respectively, without requiring extensive temporal augmentation.