image 1
image 2
image 3
image 1
image 2
image 3
image 1
image 2
image 3
| Method | F1 score |
|---|---|
| vox | 42.350 |
| point (5000 points) | 96.385 |
| point (1000 points) | 95.755 |
| point (200 points) | 86.344 |
| mesh | 87.354 |
as shown in the table above, when sampling less points in point cloud decoding method, the performance f1-score decreases, which means more points benefits the method
we visualized the features of each stage of the decoder, as shown in the picture. We can find that the decoder is gradually decoding the features to a 3d model from course to fine, from layer 0 to layer 4. Although this method is 100% reliable, we can still see the feature transformation process
