Papers by Sergei Tulyakov
LADIS: Language Disentanglement for 3D Shape Editing (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for 3D shape editing produce decoupled, local edits to 3D shapes. |
| Approach: | They propose a tool set that learns latent representations that ground language in 3D geometry. |
| Outcome: | The proposed method outperforms existing SOTA methods by 20% in edit locality and up to 6.6% in language reference resolution accuracy. |