Long-Term Temporal Convolutions for Action Recognition

TLDR

Human actions span several seconds and exhibit distinct spatio‑temporal patterns, yet existing CNN‑based methods typically learn representations from only a few frames, missing the full temporal extent. The study aims to learn video representations using long‑term temporal convolutions (LTC). The authors train LTC‑CNNs on raw pixels and optical flow, showing that high‑quality flow is crucial for accurate action models. LTC‑CNNs with longer temporal kernels outperform prior methods, achieving 92.7 % on UCF101 and 67.2 % on HMDB51, and high‑quality optical flow is essential for this performance.

Abstract

Typical human actions last several seconds and exhibit characteristic spatio-temporal structure. Recent methods attempt to capture this structure and learn action representations with convolutional neural networks. Such representations, however, are typically learned at the level of a few video frames failing to model actions at their full temporal extent. In this work we learn video representations using neural networks with long-term temporal convolutions (LTC). We demonstrate that LTC-CNN models with increased temporal extents improve the accuracy of action recognition. We also study the impact of different low-level representations, such as raw values of video pixels and optical flow vector fields and demonstrate the importance of high-quality optical flow estimation for learning accurate action models. We report state-of-the-art results on two challenging benchmarks for human action recognition UCF101 (92.7%) and HMDB51 (67.2%).