Research map / 3D vision / 3D generation
DreamDissector: Learning Disentangled Text-to-3D Generation from 2D Diffusion Priors
Zizheng Yan, Jiapeng Zhou, Fanpeng Meng, Yushuang Wu, Lingteng Qiu, Zisheng Ye, Shuguang Cui, Guanying CHEN, Xiaoguang Han
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 |
|---|---|---|---|
| disentangle a multi-object NeRF with DreamDissector instead of negative prompting | multi-object text-to-3D NeRF · Multi-object text-to-3D prompts | CLIP score (CLIP-B-16): 0.299 -> 0.316 | Improved |
| disentangle with DreamDissector instead of the composition baseline | multi-object text-to-3D NeRF · Multi-object text-to-3D prompts | CLIP score (CLIP-L-14): 0.234 -> 0.270 | Improved |
| add Deep Concept Mining to Category Score Distillation Sampling (ablation) | NeCF disentanglement · concept-gap cases (frog on water lily, beagle, octopus) | qualitative disentanglement | Improved |
| use DCM instead of SA3D for disentanglement (ablation) | NeCF disentanglement · occlusion cases (beagle, octopus) | qualitative disentanglement | Improved |
| use DCM-finetuned diffusion for DMTet refinement instead of original Stable Diffusion (ablation) | DMTet refinement · stack of pancakes (qualitative) | artifact removal | Improved |
| remove masked attention loss, first stage training or second stage training from DCM (ablation) | Deep Concept Mining · baby bunny on pancakes (qualitative) | concept separation | Hurt |
| apply TEXTure per disentangled object instead of the whole multi-object mesh | TEXTure · texture editing cases | qualitative texture quality | Improved |
| disentangle objects in very close contact such as body and clothing | DreamDissector · failure cases | qualitative disentanglement | Hurt |
Closest papers by meaning
- Let 2D Diffusion Model Know 3D-Consistency for Robust Text-to-3D Generation (ICLR 2024 · similarity 0.89)
- DreamCatalyst: Fast and High-Quality 3D Editing via Controlling Editability and Identity Preservation (ICLR 2025 · similarity 0.89)
- Disentangled 3D Scene Generation with Layout Learning (ICML 2024 · similarity 0.88)
- SceneFactor: Factored Latent 3D Diffusion for Controllable 3D Scene Generation (CVPR 2025 · similarity 0.88)
- Progressive Rendering Distillation: Adapting Stable Diffusion for Instant Text-to-Mesh Generation without 3D Data (CVPR 2025 · similarity 0.88)
- DreamTime: An Improved Optimization Strategy for Diffusion-Guided 3D Generation (ICLR 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.
