Flow-Grounded Spatial-Temporal Video Prediction from Still Images
This addresses the challenge of video prediction with minimal input for applications like video generation and simulation, representing a novel approach beyond incremental improvements.
The paper tackles the problem of generating multiple consecutive future video frames from a single still image, proposing a two-phase method that predicts multiple optical flows and then synthesizes frames, achieving favorable results in quality, diversity, and human evaluation compared to existing methods.
Existing video prediction methods mainly rely on observing multiple historical frames or focus on predicting the next one-frame. In this work, we study the problem of generating consecutive multiple future frames by observing one single still image only. We formulate the multi-frame prediction task as a multiple time step flow (multi-flow) prediction phase followed by a flow-to-frame synthesis phase. The multi-flow prediction is modeled in a variational probabilistic manner with spatial-temporal relationships learned through 3D convolutions. The flow-to-frame synthesis is modeled as a generative process in order to keep the predicted results lying closer to the manifold shape of real video sequence. Such a two-phase design prevents the model from directly looking at the high-dimensional pixel space of the frame sequence and is demonstrated to be more effective in predicting better and diverse results. Extensive experimental results on videos with different types of motion show that the proposed algorithm performs favorably against existing methods in terms of quality, diversity and human perceptual evaluation.