Abstract:Video prediction has always been a very challenging problem in video representation learning due to the complexity in spatial structure and temporal variation. However, existing methods mainly predict videos by employing language-based memory structures from the traditional Long Short-Term Memories (LSTMs) or Gated Recurrent Units (GRUs), which may not be powerful enough to model the long-term dependencies in videos, consisting of much more complex spatiotemporal dynamics than sentences. In this paper, we propose a SpatioTemporal Attention based Memory (STAM), which can efficiently improve the long-term spatiotemporal memorizing capacity by incorporating the global spatiotemporal information in videos. In the temporal domain, the proposed STAM aims to observe temporal states from a wider temporal receptive field to capture accurate global motion information. In the spatial domain, the proposed STAM aims to jointly utilize both the high-level semantic spatial state and the low-level texture spatial states to model a more reliable global spatial representation for videos. In particular, the global spatiotemporal information is extracted with the help of an Efficient SpatioTemporal Attention Gate (ESTAG), which can adaptively apply different levels of attention scores to different spatiotemporal states according to their importance. Moreover, the proposed STAM are built with 3D convolutional layers due to their advantages in modeling spatiotemporal dynamics for videos. Experimental results show that the proposed STAM can achieve state-of-the-art performance on widely used datasets by leveraging the proposed spatiotemporal representations for videos.

Unit Frame 3 : T + 2 Frame 1 : T Frame 2 : T + 1 FrameT + 1 FrameT + 2 FrameT + 3

Frame Prediction Using Recurrent Convolutional Encoder with Residual Learning

Temporal Attention Unit: Towards Efficient Spatiotemporal Predictive Learning

ASTM - an Attention Based Spatiotemporal Model for Video Prediction Using 3D Convolutional Neural Networks.

Improved 3D LSTM: A Video Prediction Approach to Long Sequence Load Forecasting

Exploring and Exploiting High-Order Spatial-Temporal Dynamics for Long-Term Frame Prediction

Unsupervised Learning of Long-Term Motion Dynamics for Videos

A Hybrid Transformer-LSTM Model With 3D Separable Convolution for Video Prediction

UNIMEMnet: Learning Long-Term Motion and Appearance Dynamics for Video Prediction with a Unified Memory Network

Predrnn: Recurrent neural networks for predictive learning using spatiotemporal lstms

Temporally-Embedded Self-Supervised Video Representation Learning

Video Prediction Recalling Long-term Motion Context via Memory Alignment Learning

Self-attention eidetic 3D-LSTM: Video prediction models for traffic flow forecasting

Triplet Attention Transformer for Spatiotemporal Predictive Learning

Inception-inspired LSTM for Next-frame Video Prediction

Recurrent Neural Network for Learning DenseDepth and Ego-Motion from Video

STAM: A SpatioTemporal Attention Based Memory for Video Prediction

Video Frame Prediction with Dual-Stream Deep Network Emphasizing Motions and Content Details.

Multi Resolution LSTM For Long Term Prediction In Neural Activity Video

PredRNN: A Recurrent Neural Network for Spatiotemporal Predictive Learning