Table 3 Quantitative results of different models on TaxiBJ.
From: 3D long time spatiotemporal convolution for complex transfer sequence prediction
Methods | \(MSE_{t=4}\) | \(MSE_{t=5}\) | \(MSE_{t=6}\) | \(MSE_{t=7}\) | \(MSE_{average}\) \(\downarrow\) |
|---|---|---|---|---|---|
ConvLSTM | 0.680 | 0.943 | 1.135 | 1.350 | 1.027 |
ST-ResNet | 0.688 | 0.939 | 1.130 | 1.288 | 1.011 |
VPN | 0.744 | 1.031 | 1.251 | 1.444 | 1.118 |
PreRNN | 0.634 | 0.934 | 1.047 | 1.263 | 0.919 |
MIM | 0.633 | 0.799 | 0.973 | 1.110 | 0.879 |
E3D-LSTM | 0.620 | 0.773 | 0.888 | 0.984 | 0.816 |
STAE | 0.521 | 0.634 | 0.701 | 0.876 | 0.683 |
STAM | 0.394 | 0.414 | 0.423 | 0.534 | 0.441 |
TCLIF | 0.490 | 0.568 | 0.622 | 0.692 | 0.592 |
STC-LIF | 0.481 | 0.548 | 0.589 | 0.630 | 0.564 |
Ours | 0.312 | 0.350 | 0.418 | 0.472 | 0.390 |
