Video Prediction on Moving MNIST
0.971SSIMSwinLSTM
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SwinLSTMComputation Time (s)=18.102025.01 | 0.971 | 14.7 | 43.1 | 2.8 | — | — | |
| MSPred NoSupsupervision=none2022.03 | 0.97 | — | 25.94 | 0.03 | — | — | |
| MSPred (ours)supervision=hierarchical2022.03 | 0.97 | — | 25.99 | 0.03 | — | — | |
| PFGNetArchitecture=Recurrent-free2026.02 | 0.967 | 15.2 | — | — | — | — | |
| IAM4VP2023.03 | 0.966 | 15.3 | — | — | 49.2 | — | |
| WaveSFNetCategory=Recurrent-free2026.03 | 0.966 | 15.8 | — | — | — | — | |
| VMRNNArchitecture=Recurrent-based2026.02 | 0.965 | 16.5 | — | — | — | — | |
| VMRNNCategory=Recurrent-based2026.03 | 0.965 | 16.5 | — | — | — | — | |
| MIMO-VPReported in author paper=true2023.03 | 0.964 | 17.7 | — | — | 51.6 | — | |
| SwinLSTMArchitecture=Recurrent-based2026.02 | 0.962 | 17.7 | — | — | — | — | |
| SwinLSTMCategory=Recurrent-based2026.03 | 0.962 | 17.7 | — | — | — | — | |
| TATCategory=Recurrent-free2026.03 | 0.96 | 17.6 | — | — | — | — | |
| TAUArchitecture=Recurrent-free2026.02 | 0.957 | 19.8 | — | — | — | — | |
| TAUCategory=Recurrent-free2026.03 | 0.957 | 19.8 | — | — | — | — | |
| MMVPArchitecture=Recurrent-free2026.02 | 0.952 | 22.2 | — | — | — | — | |
| MMVPCategory=Recurrent-free2026.03 | 0.952 | 22.2 | — | — | — | — | |
| CrevNetConference=ICLR 2020, Framework=PyTorch2022.06 | 0.949 | 22.3 | — | — | — | — | |
| CrevNetArchitecture=Recurrent-based2026.02 | 0.949 | 22.3 | — | — | — | — | |
| CrevNetCategory=Recurrent-based2026.03 | 0.949 | 22.3 | — | — | — | — | |
| SimVPConference=null, Framework=PyTorch2022.06 | 0.948 | 23.8 | — | — | — | — | |
| SimVP2023.03 | 0.948 | 23.8 | — | — | 68.9 | — | |
| SimVPArchitecture=Recurrent-free2026.02 | 0.948 | 23.8 | — | — | — | — | |
| SimVPCategory=Recurrent-free2026.03 | 0.948 | 23.8 | — | — | — | — | |
| PhyDNetConference=CVPR 2020, Framework=PyTorch2022.06 | 0.947 | 24.4 | — | — | — | — | |
| PhyDNet2023.03 | 0.947 | 24.4 | — | — | 70.3 | — | |
| PhyDNetArchitecture=Recurrent-based2026.02 | 0.947 | 24.4 | — | — | — | — | |
| PhyDNetCategory=Recurrent-based2026.03 | 0.947 | 24.4 | — | — | — | — | |
| MAUArchitecture=Recurrent-based2026.02 | 0.937 | 27.6 | — | — | — | — | |
| MAUCategory=Recurrent-based2026.03 | 0.937 | 27.6 | — | — | — | — | |
| Proposed ModelComputation Time (s)=14.102025.01 | 0.935 | 43.5 | 22.5 | 4.8 | — | — | |
| LMCArchitecture=Recurrent-based2026.02 | 0.924 | 41.5 | — | — | — | — | |
| E3D-LSTMConference=ICLR 2018, Framework=Tensorflow2022.06 | 0.92 | 41.3 | — | — | — | — | |
| E3D-LSTM2023.03 | 0.92 | 41.3 | — | — | 86.4 | — | |
| PhyDNet2022.03 | 0.915 | — | 20.43 | 0.054 | — | — | |
| PredRNN++2022.03 | 0.911 | — | 20.2 | 0.055 | — | — | |
| MIMConference=CVPR 2018, Framework=Tensorflow2022.06 | 0.91 | 44.2 | — | — | — | — | |
| MIM2023.03 | 0.91 | 44.2 | — | — | 101.1 | — | |
| MIMArchitecture=Recurrent-based2026.02 | 0.91 | 44.2 | — | — | — | — | |
| E3D-LSTMArchitecture=Recurrent-based2026.02 | 0.91 | 41.3 | — | — | — | — | |
| MIMCategory=Recurrent-based2026.03 | 0.91 | 44.2 | — | — | — | — | |
| MAUComputation Time (s)=19.802025.01 | 0.91 | 48 | 19.6 | 5.98 | — | — | |
| SAVPComputation Time (s)=18.902025.01 | 0.909 | 28.61 | 18.2 | 6.13 | — | — | |
| SVG-LPtype=Learned Prior2022.03 | 0.907 | — | 20.36 | 0.115 | — | — | |
| SVG-Dettype=Deterministic2022.03 | 0.9 | — | 20.31 | 0.114 | — | — | |
| CausalLSTMConference=ICML 2018, Framework=Tensorflow2022.06 | 0.898 | 46.5 | — | — | — | — | |
| Causal LSTM2023.03 | 0.898 | 46.5 | — | — | 106.8 | — | |
| CausalLSTMArchitecture=Recurrent-based2026.02 | 0.898 | 46.5 | — | — | — | — | |
| Earthformer#Param. (M)=7.6, GFLOPS=34.02022.07 | 0.8961 | 41.79 | — | — | 92.78 | — | |
| TrajGRU2022.03 | 0.895 | — | 20.02 | 0.075 | — | — | |
| PredRNN-V2Conference=Arxiv 2021, Framework=PyTorch2022.06 | 0.891 | 48.4 | — | — | — | — | |
| STAUComputation Time (s)=16.002025.01 | 0.885 | 25.41 | 18.5 | 5.3 | — | — | |
| PredRNN#Param. (M)=23.8, GFLOPS=232.02022.07 | 0.8831 | 52.07 | — | — | 108.9 | — | |
| Earthformer w/o global#Param. (M)=6.6, GFLOPS=33.72022.07 | 0.8825 | 46.91 | — | — | 101.5 | — | |
| E3D-LSTM#Param. (M)=12.9, GFLOPS=302.02022.07 | 0.8821 | 55.31 | — | — | 101.6 | — | |
| PredRNNComputation Time (s)=59.002025.01 | 0.88 | 75.1 | 18.1 | 6.9 | — | — | |
| PredRNN-V2Computation Time (s)=51.502025.01 | 0.879 | 74.2 | 17.9 | 6.8 | — | — | |
| SVGComputation Time (s)=19.902025.01 | 0.872 | 81 | 17.7 | 8.2 | — | — | |
| VPNArchitecture=Recurrent-based2026.02 | 0.87 | 64.1 | — | — | — | — | |
| PredRNNConference=NIPS 2017, Framework=PyTorch2022.06 | 0.867 | 56.8 | — | — | — | — | |
| PredRNN2023.03 | 0.867 | 56.8 | — | — | 126.1 | — | |
| PredRNNArchitecture=Recurrent-based2026.02 | 0.867 | 56.8 | — | — | — | — | |
| PredRNNCategory=Recurrent-based2026.03 | 0.867 | 56.8 | — | — | — | — | |
| ConvLSTM#Param. (M)=14.0, GFLOPS=30.12022.07 | 0.8477 | 62.04 | — | — | 126.9 | — | |
| PhyDnetComputation Time (s)=20.602025.01 | 0.837 | 62.5 | 19.3 | 14.8 | — | — | |
| PhyDNet#Param. (M)=3.1, GFLOPS=15.32022.07 | 0.835 | 58.7 | — | — | 124.1 | — | |
| ConvLSTM2022.03 | 0.833 | — | 17.22 | 0.144 | — | — | |
| FRNNArchitecture=Recurrent-based2026.02 | 0.813 | 69.7 | — | — | — | — | |
| FRNNCategory=Recurrent-based2026.03 | 0.813 | 69.7 | — | — | — | — | |
| MMVPComputation Time (s)=30.102025.01 | 0.802 | 89.1 | 15.3 | 19.3 | — | — | |
| CrevNetComputation Time (s)=23.802025.01 | 0.76 | 97.8 | 17 | 20 | — | — | |
| LMCNetComputation Time (s)=23.002025.01 | 0.76 | 92 | 17.5 | 17.1 | — | — | |
| Eidetic 3D LSTMComputation Time (s)=31.002025.01 | 0.749 | 90.1 | 17.5 | 17 | — | — | |
| Rainformer#Param. (M)=19.2, GFLOPS=1.22022.07 | 0.7301 | 85.83 | — | — | 189.2 | — | |
| DFNArchitecture=Recurrent-based2026.02 | 0.726 | 89 | — | — | — | — | |
| ConvLSTMConference=NIPS 2015, Framework=PyTorch2022.06 | 0.707 | 103.3 | — | — | — | — | |
| ConvLSTM2023.03 | 0.707 | 103.3 | — | — | 182.9 | — | |
| ConvLSTMArchitecture=Recurrent-based2026.02 | 0.707 | 103.3 | — | — | — | — | |
| ConvLSTMCategory=Recurrent-based2026.03 | 0.707 | 103.3 | — | — | — | — | |
| UNet#Param. (M)=16.6, GFLOPS=0.92022.07 | 0.617 | 110.4 | — | — | 249.4 | — | |
| TCVBM2025.10 | 0.589 | — | 10.71 | 0.26 | — | 45.32 | |
| BM2025.10 | 0.582 | — | 10.627 | 0.268 | — | 48.54 | |
| DDPM2025.10 | 0.53 | — | 10.333 | 0.383 | — | 250.52 | |
| DDIM2025.10 | 0.513 | — | 10.205 | 0.402 | — | 335.51 |