Interactive3D🪄: Create What You Want by Interactive 3D Generation

CVPR 2024

Shaocong Dong*1, Lihe Ding*2,4, Zhanpeng Huang3, Zibin Wang3, Tianfan Xue†2, Dan Xu†1
1HKUST, 2CUHK, 3SenseTime Research, 4Shanghai AI Lab
arXiv Video Code


Interactive3D🪄 achieves controllable and high-quality 3D generation.

Method

MY ALT TEXT

The overall architecture: Interactive3D grants users precise control over the generative process through extensive 3D interaction capabilities. Interactive3D is constructed in two cascading stages, utilizing distinct 3D representations. The first stage employs Gaussian Splatting for direct user interaction, allowing modifications and guidance of the generative direction at any intermediate step through (i) Adding and Removing components, (ii) Deformable and Rigid Dragging, (iii) Geometric Transformations, and (iv) Semantic Editing. Subsequently, the Gaussian splats are transformed into InstantNGP. We introduce a novel (v) Interactive Hash Refinement module to further add details and extract the geometry in the second stage.

Rigid Dragging

Using rigid dragging to open a T-rex's mouth.
"A tyrannosaurus rex".

Part Combination

Using part combination to assemble your gundam.
"A gundam robot holding a sword with wings".

Hash Refinement

Using hash refinement to improve the quality.
"A gundam robot holding a sword with wings".

Semantic Editing

Using semantic editing to change your gundam.
"A gundam robot holding a ice sword with wings made of ice".

Deformable Dragging

Using deformable dragging to create banana whale.
"A whale that looks like a banana".

Geometric Transformation

Using geometric transformation to open a pokeball.
"A baby flying dragon in an open pokeball".

BibTeX

@article{dong2024interactive3d,
  title={Interactive3D: Create What You Want by Interactive 3D Generation},
  author={Dong, Shaocong and Ding, Lihe and Huang, Zhanpeng and Wang, Zibin and Xue, Tianfan and Xu, Dan},
  journal={arXiv preprint arXiv:2404.16510},
  year={2024}
}