Research map / 3D vision / 3D generation
TeRA: Rethinking Text-guided Realistic 3D Avatar Generation
Yanwen Wang, Yiyu Zhuang, Jiawei Zhang, Li Wang, Yifei Zeng, Xun Cao, Xinxin Zuo, Hao Zhu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use TeRA feedforward latent diffusion instead of SDS optimization (HumanNorm) | HumanNorm · Text-to-3D avatar generation | generation time: 4.0h -> 12s | Improved |
| use TeRA instead of HumanNorm: user-study realism | HumanNorm · Text-to-3D avatar generation | user study realism: 3.04 -> 4.35 | Improved |
| use TeRA instead of HumanNorm: user-study text consistency | HumanNorm · Text-to-3D avatar generation | user study text consistency: 3.79 -> 4.54 | Improved |
| use TeRA instead of HumanNorm: user-study visual quality | HumanNorm · Text-to-3D avatar generation | user study visual quality: 3.01 -> 4.33 | Improved |
| use TeRA instead of X-Oscar: CLIP score | X-Oscar · Text-to-3D avatar generation | CLIP Score: 32.46 -> 30.17 | Hurt |
| use TeRA instead of HumanNorm: VQA score | HumanNorm · Text-to-3D avatar generation | VQA Score: 0.72 -> 0.82 | Improved |
| use TeRA instead of HumanGaussian: VQA score | HumanGaussian · Text-to-3D avatar generation | VQA Score: 0.82 -> 0.82 | No clear effect |
| raise structured latent resolution from 128x128 to 256x256 (ablation) | TeRA · TeRA latent space ablation | qualitative artifacts (Fig.) | Improved |
| use diffusion inpainting on structured latent instead of direct feature swapping for virtual try-on (ablation) | TeRA · virtual try-on editing | qualitative artifacts (Fig.) | Improved |
| add distillation module linking diffusion model to VAE encoder (instead of direct connection) (ablation) | TeRA · TeRA training | generation quality | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- HumanGaussian — HumanGaussian: Text-Driven 3D Human Generation with Gaussian Splatting (CVPR 2024) · better on 4 of 6 shared results
- HumanNorm — HumanNorm: Learning Normal Diffusion Model for High-quality and Realistic 3D Human Generation (CVPR 2024) · better on 5 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- DreamAvatar: Text-and-Shape Guided 3D Human Avatar Generation via Diffusion Models (CVPR 2024 · similarity 0.89)
- AvatarVerse: High-Quality & Stable 3D Avatar Creation from Text and Pose (AAAI 2024 · similarity 0.89)
- DiffusionGAN3D: Boosting Text-guided 3D Generation and Domain Adaptation by Combining 3D GANs and Diffusion Priors (CVPR 2024 · similarity 0.89)
- En3D: An Enhanced Generative Model for Sculpting 3D Humans from 2D Synthetic Data (CVPR 2024 · similarity 0.89)
- X-Oscar: A Progressive Framework for High-quality Text-guided 3D Animatable Avatar Generation (ICML 2024 · similarity 0.88)
- Morphable Diffusion: 3D-Consistent Diffusion for Single-image Avatar Creation (CVPR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
