Table 6 Performance comparison of various methods on NewCollege and CityCentre.
From: A visual SLAM loop closure detection method based on lightweight siamese capsule network
| Â | Methods | Accuracy | Time(s) |
|---|---|---|---|
NewCollege | OURS | 0.936 | 0.074 |
SqueezeNet | 0.933 | 0.044 | |
GoogleNet | 0.664 | 0.063 | |
DarkNet | 0.801 | 0.047 | |
PCANet | 0.738 | 0.012 | |
FLCNN | 0.794 | 0.045 | |
VGG | 0.839 | 0.077 | |
MobileNet | 0.725 | 0.063 | |
OURS(without pruning) | 0.951 | 0.107 | |
LSGD | 0.936 | / | |
CityCentre | OURS | 0.947 | 0.074 |
SqueezeNet | 0.940 | 0.044 | |
GoogleNet | 0.742 | 0.063 | |
DarkNet | 0.853 | 0.047 | |
PCANet | 0.813 | 0.012 | |
FLCNN | / | 0.045 | |
VGG | 0.861 | 0.077 | |
MobileNet | 0.791 | 0.063 | |
OURS(without pruning) | 0.964 | 0.107 | |
LSGD | 0.974 | / |
