Research map / 3D vision / 3D generation
Learn to Optimize Denoising Scores: A Unified and Improved Diffusion Prior for 3D Generation
Xiaofeng Yang, Yiwen Chen, Cheng Chen, Chi Zhang, Yi Xu, Xulei Yang, Fayao Liu, Guosheng Lin
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 LODS (learnable null embedding) instead of VSD loss in ProlificDreamer | text-to-3D with Stable Diffusion · T3Bench (all categories) | Average: 43.3 -> 47.3 | Improved |
| use LODS (LoRA) instead of VSD loss | text-to-3D with Stable Diffusion · T3Bench (all categories) | Average: 43.3 -> 45.4 | Improved |
| use embedding-based LODS instead of LoRA-based LODS (ablation) | LODS · T3Bench (all categories) | Average: 45.4 -> 47.3 | Improved |
| replace SDS loss with LODS inside Magic123 for the Stable Diffusion prior | Magic123 · NeRF4 image-to-3D | PSNR: 24.62 -> 24.83 | No clear effect |
| replace SDS loss with LODS inside Magic123 | Magic123 · RealFusion15 image-to-3D | LPIPS: 0.10 -> 0.09 | No clear effect |
| start with low classifier-free guidance weight (30) in embedding LODS (ablation) | LODS (Emb) · text-to-3D qualitative | texture detail | Hurt |
| increase initial CFG weight beyond 1000 up to infinity (ablation) | LODS (Emb) · text-to-3D qualitative | 3D quality | Hurt |
| combine embedding-based and LoRA-based LODS (ablation) | LODS · 3D generation | generation quality | No clear effect |
| gradually decrease the CFG weight during optimization (ablation) | SDS loss · text-to-3D | generation quality | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Magic123 — Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors (ICLR 2024) · better on 3 of 4 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
- DreamTime: An Improved Optimization Strategy for Diffusion-Guided 3D Generation (ICLR 2024 · similarity 0.90)
- Score Distillation via Reparametrized DDIM (NeurIPS 2024 · similarity 0.90)
- Denoising Diffusion Bridge Models (ICLR 2024 · similarity 0.90)
- A Simple Early Exiting Framework for Accelerated Sampling in Diffusion Models (ICML 2024 · similarity 0.90)
- Denoising Task Difficulty-based Curriculum for Training Diffusion Models (ICLR 2025 · similarity 0.89)
- Evaluating the design space of diffusion-based generative models (NeurIPS 2024 · similarity 0.89)
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.
