3D Object Reconstruction on ShapeNet 12 (test)
0.99Chair ScoreNKRR (Ours)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| NKRR (Ours)Input Point Cloud Size=10k, Noise Level (Gaussian)=0.005, Evaluation Protocol=Nyström KRR in feature space, Optimizer=Adam, Training Epochs=100, Inducing Points=500, Feature Dimensions=322023.11 | 0.99 | 0.96 | 0.99 | 0.98 | |
| SACInput Point Cloud Size=10k, Noise Level (Gaussian)=0.005, Evaluation Protocol=finetuning strategy, Optimizer=Adam, Training Epochs=1002023.11 | 0.98 | 0.94 | 0.92 | 0.95 | |
| PocoInput Point Cloud Size=10k, Noise Level (Gaussian)=0.005, Evaluation Protocol=deep data prior baseline, Optimizer=Adam, Training Epochs=1002023.11 | 0.98 | 0.95 | 0.98 | 0.97 | |
| ConvInput Point Cloud Size=10k, Noise Level (Gaussian)=0.005, Evaluation Protocol=deep convolutional occupancy models, Optimizer=Adam, Training Epochs=1002023.11 | 0.95 | 0.92 | 0.98 | 0.95 | |
| NPInput Point Cloud Size=10k, Noise Level (Gaussian)=0.005, Evaluation Protocol=data prior free overfitting, Optimizer=Adam, Training Epochs=1002023.11 | 0.87 | 0.83 | 0.8 | 0.83 |