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.
|Title of host publication
|Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings
|Martial Hebert, Vittorio Ferrari, Cristian Sminchisescu, Yair Weiss
|Number of pages
|Published - 2018
|15th European Conference on Computer Vision, ECCV 2018 - Munich, Germany
Duration: 2018 Sept 8 → 2018 Sept 14
|Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
|15th European Conference on Computer Vision, ECCV 2018
|18/9/8 → 18/9/14
Bibliographical notePublisher Copyright:
© Springer Nature Switzerland AG 2018.
All Science Journal Classification (ASJC) codes
- Theoretical Computer Science
- General Computer Science