Below we compare our results against ChangeIt3D over the semantic shape editing task. We show guidance-output shape pairs (guidance shapes are depicted on the left), all represented as point clouds, as ChangeIt3D operates over point clouds. Note that our guidance shapes are extracted from Shape-E's encoding, allowing for better emphasizing the effect of the edits (vs. showing the input shapes directly).
| ChangeIt3D | Ours | |
|---|---|---|
|
|
|
`
|
| | ||
|
It has no strecher* between the feet |
|
`
|
| *A stretcher is a horizontal support element (typically located on a table, chair or other item of furniture). | ||
| Observe how our method allows for removing the stretchers from the bottom of the feet while preserving their structure, while ChangeIt3D instead connects the feet together. | ||
| | ||
| ChangeIt3D | Ours | |
|
|
|
|
| | ||
|
|
|
|
| | ||
| ChangeIt3D | Ours | |
|
|
|
|
| | ||
|
|
|
|