Comparison of model sizes, computational complexities (in FLOPs), and peak memory requirements.
| Method | Model Size | Total FLOPs | Peak Memory | Runtime | |
|---|---|---|---|---|---|
| Unsupervised | PGR [33] | <100 B | – | 11815 MB (x88) | 7.41 s |
| GPSR[59] | <100 B | 10 G (x1) | 134 MB (x1) | 13.20 s | |
| Supervised | Points2surf [14] | 22 MB (x3.93) | 5301 G (x62) | 7805 MB (x8.71) | 46.49 s |
| SR+PCPNet [19] | 87 MB (x15.53) | 2688 G (x31) | 4967 MB (x5.54) | 8.97 s | |
| Neural-Pull [3] | 15 MB (x2.67) | 85486 G (x994) | * | 362 s | |
| POCO [5] | 133 MB (x23.75) | 2370 G (x28) | 33276 MB (x37.13) | 39.53 s | |
| NKSR [24] | 57 MB (x10.18) | 340 G (x4) | 3409 MB (x3.80) | 1.70 s | |
| Geo-UDF [41] | 3.2 MB (x0.57) | 690 G (x8.02) | 5742 MB (x6.41) | 65.63 s | |
| CAP-UDF [55] | 2.0 MB (x0.35) | 256491 G (x2982) | * | 947 s | |
| LevelSetUDF [56] | 1.8 MB (x0.32) | 330702 G (x3845) | * | 789 s | |
| LPSR (Ours) | 5.6 MB (x1) | 86 G (x1) | 896 MB (x1) | 21.17 s | |
| Method | Model Size | Total FLOPs | Peak Memory | Runtime | |
|---|---|---|---|---|---|
| Unsupervised | PGR [ | – | 11815 MB (x88) | 7.41 s | |
| GPSR[ | 13.20 s | ||||
| Supervised | Points2surf [ | 22 MB (x3.93) | 5301 G (x62) | 7805 MB (x8.71) | 46.49 s |
| SR+PCPNet [ | 87 MB (x15.53) | 2688 G (x31) | 4967 MB (x5.54) | 8.97 s | |
| Neural-Pull [ | 15 MB (x2.67) | 85486 G (x994) | 362 s | ||
| POCO [ | 133 MB (x23.75) | 2370 G (x28) | 33276 MB (x37.13) | 39.53 s | |
| NKSR [ | 57 MB (x10.18) | 340 G (x4) | 3409 MB (x3.80) | ||
| Geo-UDF [ | 3.2 MB (x0.57) | 690 G (x8.02) | 5742 MB (x6.41) | 65.63 s | |
| CAP-UDF [ | 2.0 MB (x0.35) | 256491 G (x2982) | 947 s | ||
| LevelSetUDF [ | 330702 G (x3845) | 789 s | |||
| LPSR (Ours) | 5.6 MB (x1) | 21.17 s | |||
Neural-Pull, CAP-UDF and LevelSetUDF requires online training for each query point cloud. Its peak memory is GPU specific.