Stochastic Variational Video Prediction

Predicting the future in real-world settings, particularly from raw sensory observations such as images, is exceptionally challenging. Real-world events can be stochastic and unpredictable, and the high dimensionality and complexity of natural images requires the predictive model to build an intricate understanding of the natural world. Many existing methods tackle this problem by making simplifying assumptions about the environment. One common assumption is that the outcome is deterministic and there is only one plausible future. This can lead to low-quality predictions in real-world settings with stochastic dynamics. In this paper, we develop a stochastic variational video prediction (SV2P) method that predicts a different possible future for each sample of its latent variables. To the best of our knowledge, our model is the first to provide effective stochastic multi-frame prediction for real-world video. We demonstrate the capability of the proposed method in predicting detailed future frames of videos on multiple real-world datasets, both action-free and action-conditioned. We find that our proposed method produces substantially improved video predictions when compared to the same model without stochasticity, and to other stochastic video prediction methods. Our SV2P implementation will be open sourced upon publication.

PDF Abstract ICLR 2018 PDF ICLR 2018 Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Video Generation BAIR Robot Pushing SV2P (from FVD) FVD score 262.5 # 24
Cond 2 # 12
Pred 14 # 2
Train 14 # 11
Video Generation BAIR Robot Pushing SV2P (from SRVP) FVD score 965±17 # 29
Cond 2 # 12
SSIM 0.8169±0.0086 # 7
PSNR 20.39±0.27 # 2
LPIPS 0.0912±0.0053 # 2
Pred 28 # 19
Train 12 # 17
Video Prediction KTH SV2P time-variant (from Grid-keypoints) LPIPS 0.232 # 15
PSNR 25.87 # 24
FVD 209.5 # 5
SSIM 0.782 # 23
Cond 10 # 1
Pred 40 # 22
Params (M) 8.3 # 4
Train 10 # 1
Video Prediction KTH SV2P time-invariant (from Grid-keypoints) LPIPS 0.260 # 16
PSNR 25.70 # 25
FVD 253.5 # 8
SSIM 0.772 # 24
Cond 10 # 1
Pred 40 # 22
Params (M) 8.3 # 4
Train 10 # 1
Video Prediction KTH SV2P (from SRVP) LPIPS 0.2049±0.0053 # 13
PSNR 28.19±0.31 # 8
FVD 636 ± 1 # 12
SSIM 0.838 # 13
Cond 10 # 1
Pred 30 # 17
Train 10 # 1

Methods


No methods listed for this paper. Add relevant methods here