Table 1 Performance evaluation of domain–domain translation models
From: Modelling intra-muscular contraction dynamics using in silico to in vivo domain translation
Direction | Translation model | Frames | Spatial features | Temporal features | Translated content consistency | |||
|---|---|---|---|---|---|---|---|---|
DBhat \(\downarrow \) | Corr \(\uparrow \) | SSI \(\uparrow \) | Corr Psd \(\uparrow \) | xCorr \(\downarrow \) | Lag, ms \(\downarrow \) | |||
Sim to Exp | No translation | – | 0.33 (0.11) | 0.54 (0.26) | 0.32 (0.18) | 0.96 (0.01) | – | – |
-”- | 2D cycleGAN | 1 | 0.43 (0.10) | 0.32 (0.19) | 0.49 (0.12) | 0.90 (0.02) | 0.61 (0.09) | − 2.0 (64.6) |
-”- | Recycle GAN | 3 | 0.39 (0.11) | 0.39 (0.20) | 0.65 (0.14) | 0.96 (0.01) | 0.51 (0.07) | − 1.9 (70.8) |
-”- | 3D cycleGAN default | 32 | 0.25 (0.09) | 0.67 (0.21) | 0.49 (0.08) | 0.90 (0.02) | 0.89 (0.04) | 1.3 (2.4) |
-”- | 3D cycleGAN (Stride=1) | 32 | 0.27 (0.12) | 0.65 (0.25) | 0.63 (0.09) | 0.97 (0.01) | 0.93 (0.03) | − 1.7 (2.2) |
-”- | 3D cycleGAN (Stride 1 + Noise) | 32 | 0.25 (0.10) | 0.69 (0.23) | 0.65 (0.10) | 0.97 (0.01) | 0.93 (0.03) | 1.4 (2.2) |
Exp to Exp | No translation | – | 0.26 (0.11) | 0.66 (0.25) | 0.74 (0.12) | 0.98 (0.01) | – | – |