TABLE II.
Ablation Study on Our MDPET in Terms of Motion Estimation. ± LSTM Means MDPET With or Without BiConvLSTM and ± GAN Means MDPET With or Without Adversarial Learning.
NMAE | Gate 1 | Gate 2 | Gate 3 | Gate 4 | Gate 5 | Gate 6 | CPU sec |
---|---|---|---|---|---|---|---|
Ours-LSTM-GAN | 0.1283 | 0.0849 | 0.0660 | - | 0.0701 | 0.1010 | 36 |
Ours+LSTM-GAN | 0.1116* | 0.0762* | 0.0588* | - | 0.0623* | 0.0920* | 59 |
Ours+LSTM+GAN | 0.1098* | 0.0749* | 0.0582* | - | 0.0619* | 0.0908* | 59 |
Means the Difference to the Baseline (1st Row) Are Significant at p < 0.05