Decomposing Motion and Content for Natural Video Sequence Prediction

Abstract

We propose a deep neural network for the prediction of future frames in natural video sequences. To effectively handle complex evolution of pixels in videos, we propose to decompose the motion and content, two key components generating dynamics in videos. Our model is built upon the Encoder-Decoder Convolutional Neural Network and Convolutional LSTM for pixel-level prediction, which independently capture the spatial layout of an image and the corresponding temporal dynamics. By independently modeling motion and content, predicting the next frame reduces to converting the extracted content features into the next frame content by the identified motion features, which simplifies the task of prediction. Our model is end-to-end trainable over multiple time steps, and naturally learns to decompose motion and content without separate training. We evaluate the proposed network architecture on human activity videos using KTH, Weizmann action, and UCF-101 datasets. We show state-of-the-art performance in comparison to recent approaches. To the best of our knowledge, this is the first end-to-end trainable network architecture with motion and content separation to model the spatiotemporal dynamics for pixel-level future prediction in natural videos.

International Conference on Learning Representations (ICLR) 2017

Cited by in corpus (35)

• Deep Learning for Precipitation Nowcasting: A Benchmark and A New Model
• Recent Advances in Autoencoder-Based Representation Learning
• A Review on Generative Adversarial Networks: Algorithms, Theory, and Applications
• Detecting Adversarial Attacks on Neural Network Policies with Visual Foresight
• Dual Contrastive Learning for Spatio-temporal Representation
• Learning Disentangled Representations with Reference-Based Variational Autoencoders
• Dynamics Transfer GAN: Generating Video by Transferring Arbitrary Temporal Dynamics from a Source Video to a Single Target Image
• Disentangled State Space Representations
• Prediction Under Uncertainty with Error-Encoding Networks
• Video Generation from Single Semantic Label Map
• TubeTK: Adopting Tubes to Track Multi-Object in a One-Step Training Model
• Few-shot Video-to-Video Synthesis
• S3VAE: Self-Supervised Sequential VAE for Representation Disentanglement and Data Generation
• Anomaly Detection using Deep Reconstruction and Forecasting for Autonomous Systems
• Order Matters: Shuffling Sequence Generation for Video Prediction
• Self-Supervised Learning via multi-Transformation Classification for Action Recognition
• Learning World Transition Model for Socially Aware Robot Navigation
• Review of Video Predictive Understanding: Early Action Recognition and Future Action Prediction
• Disentangling Motion, Foreground and Background Features in Videos
• Predicting tongue motion in unlabeled ultrasound videos using convolutional LSTM neural network
• Future Frame Prediction for Robot-assisted Surgery
• Inserting Videos into Videos
• Motion Selective Prediction for Video Frame Synthesis
• Causal Future Prediction in a Minkowski Space-Time
• DeepLandscape: Adversarial Modeling of Landscape Video
• Sound2Sight: Generating Visual Dynamics from Sound and Context
• Exploring Self-Attention for Visual Odometry
• Hierarchical Video Generation for Complex Data
• Towards Purely Unsupervised Disentanglement of Appearance and Shape for Person Images Generation
• Learning Semantic-Aware Dynamics for Video Prediction
• FREGAN : an application of generative adversarial networks in enhancing the frame rate of videos
• From Single to Multiple: Leveraging Multi-level Prediction Spaces for Video Forecasting
• Cubic LSTMs for Video Prediction
• Dual-MTGAN: Stochastic and Deterministic Motion Transfer for Image-to-Video Synthesis
• Self-Supervised Equivariant Scene Synthesis from Video