Persistence diagram reconstruction on DONUT held-out (test)
16.02W2 ErrorPoint-MAE
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Point-MAEEncoder training protocol=Pretrained-frozen, Feature selection=Combined features from all transformer blocks (C)2026.04 | 16.02 | 9.838 | 1.214 | |
| Point-BERTEncoder training protocol=Pretrained-frozen, Feature selection=Last transformer block (L)2026.04 | 16.18 | 9.901 | 1.371 | |
| DGCNNEncoder training protocol=Encoders trained end-to-end2026.04 | 16.62 | 10.15 | 1.213 | |
| PCP-MAEEncoder training protocol=Pretrained-frozen, Feature selection=Last transformer block (L)2026.04 | 17.18 | 9.95 | 1.333 | |
| Point-MAEEncoder training protocol=Pretrained-frozen, Feature selection=Last transformer block (L)2026.04 | 17.24 | 9.917 | 1.107 | |
| Point-BERTEncoder training protocol=Pretrained-frozen, Feature selection=Combined features from all transformer blocks (C)2026.04 | 17.33 | 9.918 | 1.584 | |
| PointGPTEncoder training protocol=Pretrained-frozen, Feature selection=Combined features from all transformer blocks (C)2026.04 | 17.59 | 10.08 | 1.289 | |
| PointGPTEncoder training protocol=Pretrained-frozen, Feature selection=Last transformer block (L)2026.04 | 17.86 | 10.03 | 1.192 | |
| PCP-MAEEncoder training protocol=Pretrained-frozen, Feature selection=Combined features from all transformer blocks (C)2026.04 | 18.73 | 9.933 | 2.654 | |
| PointNetEncoder training protocol=Encoders trained end-to-end2026.04 | 24.85 | 10.61 | 1.442 | |
| PointNet++Encoder training protocol=Encoders trained end-to-end2026.04 | 39.37 | 13.64 | 10.39 |