Lune

ICLR2025顶会

RecDreamer: Consistent Text-to-3D Generation via Uniform Score Distillation

Chenxi Zheng, Yihong Lin, Bangzhen Liu, Xuemiao Xu, Yongwei Nie, Shengfeng He

摘要

Current text-to-3D generation methods based on score distillation often suffer from geometric inconsistencies, leading to repeated patterns across different poses of 3D assets. This issue, known as the Multi-Face Janus problem, arises because existing methods struggle to maintain consistency across varying poses and are biased toward a canonical pose. While recent work has improved pose control and approximation, these efforts are still limited by this inherent bias, which skews the guidance during generation. To address this, we propose a solution called RecDreamer, which reshapes the underlying data distribution to achieve a more consistent pose representation. The core idea behind our method is to rectify the prior distribution, ensuring that pose variation is uniformly distributed rather than biased toward a canonical form. By modifying the prescribed distribution through an auxiliary function, we can reconstruct the density of the distribution to ensure compliance with specific marginal constraints. In particular, we ensure that the marginal distribution of poses follows a uniform distribution, thereby eliminating the biases introduced by the prior knowledge. We incorporate this rectified data distribution into existing score distillation algorithms, a process we refer to as uniform score distillation. To efficiently compute the posterior distribution required for the auxiliary function, RecDreamer introduces a training-free classifier that estimates pose categories in a plug-and-play manner. Additionally, we utilize various approximation techniques for noisy states, significantly improving system performance. Our experimental results demonstrate that RecDreamer effectively mitigates the Multi-Face Janus problem, leading to more consistent 3D asset generation across different poses.

问 Lune

几分钟读透这篇 Paper,不用再花一下午。

Lune 能读完全文,也能读引用它的顶会 Paper,讲清它的结论、证据和局限,每一点都注明出自哪篇 Paper。不用硬啃,也不靠猜。

已有 5 篇顶会 Paper 引用了它。别等审稿人问起,先弄清它们发现了什么。

可以从这些问题问起

Lune 可以查阅

全文 · 引用它的顶会 Paper · 相关的 Paper

获取答案

免费开始,无需绑卡

脉络

  • High-Resolution Image Synthesis with Latent Diffusion Models 引用了 Denoising Diffusion Probabilistic Models
  • Progressive Rendering Distillation: Adapting Stable Diffusion for Instant Text-to-Mesh Generation without 3D Data 引用了 LoRA: Low-Rank Adaptation of Large Language Models
  • Progressive Rendering Distillation: Adapting Stable Diffusion for Instant Text-to-Mesh Generation without 3D Data 引用了 High-Resolution Image Synthesis with Latent Diffusion Models
  • Progressive Rendering Distillation: Adapting Stable Diffusion for Instant Text-to-Mesh Generation without 3D Data 引用了 3D Gaussian Splatting for Real-Time Radiance Field Rendering
  • Progressive Rendering Distillation: Adapting Stable Diffusion for Instant Text-to-Mesh Generation without 3D Data 引用了 Learning Transferable Visual Models From Natural Language Supervision
  • Track, Inpaint, Resplat: Subject-driven 3D and 4D Generation with Progressive Texture Infilling 引用了 LoRA: Low-Rank Adaptation of Large Language Models
  • Track, Inpaint, Resplat: Subject-driven 3D and 4D Generation with Progressive Texture Infilling 引用了 High-Resolution Image Synthesis with Latent Diffusion Models
  • Track, Inpaint, Resplat: Subject-driven 3D and 4D Generation with Progressive Texture Infilling 引用了 3D Gaussian Splatting for Real-Time Radiance Field Rendering
  • Track, Inpaint, Resplat: Subject-driven 3D and 4D Generation with Progressive Texture Infilling 引用了 Denoising Diffusion Probabilistic Models
  • Seeing 3D Through 2D Lenses: 3D Few-Shot Class-Incremental Learning via Cross-Modal Geometric Rectification 引用了 Learning Transferable Visual Models From Natural Language Supervision
  • Think-Then-Generate: Structural Chain-of-Thought Reasoning for Consistent 3D Generation 引用了 High-Resolution Image Synthesis with Latent Diffusion Models
  • Think-Then-Generate: Structural Chain-of-Thought Reasoning for Consistent 3D Generation 引用了 3D Gaussian Splatting for Real-Time Radiance Field Rendering
  • Think-Then-Generate: Structural Chain-of-Thought Reasoning for Consistent 3D Generation 引用了 Learning Transferable Visual Models From Natural Language Supervision
它引用的引用它的从左到右按年份排列,引用越多,位置越高。

引用它的顶会 Paper5

  1. ViewSRD: 3D Visual Grounding Via Structured Multi-View DecompositionRonggang Huang, Haoxin Yang, Yan Cai, Xuemiao Xu 等ICCV 2025 · 2 次引用
  2. Track, Inpaint, Resplat: Subject-driven 3D and 4D Generation with Progressive Texture InfillingShuhong Zheng, Ashkan Mirzaei, Igor GilitschenskiNeurIPS 2025 · 2 次引用
  3. Seeing 3D Through 2D Lenses: 3D Few-Shot Class-Incremental Learning via Cross-Modal Geometric RectificationTuo Xiang, Xuemiao Xu, Bangzhen Liu, Jinyi Li 等ICCV 2025 · 1 次引用
  4. Progressive Rendering Distillation: Adapting Stable Diffusion for Instant Text-to-Mesh Generation without 3D DataZhiyuan Ma, Xinyue Liang, Rongyuan Wu, Xiangyu Zhu 等CVPR 2025
  5. Think-Then-Generate: Structural Chain-of-Thought Reasoning for Consistent 3D GenerationXinyue Liu, Jin Liu, Hongbo Wang, Ran He 等CVPR 2026

它引用的 Paper38

  1. Learning Transferable Visual Models From Natural Language SupervisionAlec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh 等ICML 2021 · 47,906 次引用
  2. Denoising Diffusion Probabilistic ModelsJonathan Ho, Ajay Jain, Pieter AbbeelNeurIPS 2020 · 35,902 次引用
  3. LoRA: Low-Rank Adaptation of Large Language ModelsEdward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu 等ICLR 2022 · 18,833 次引用
  4. High-Resolution Image Synthesis with Latent Diffusion ModelsRobin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser 等CVPR 2022 · 13,123 次引用
  5. 3D Gaussian Splatting for Real-Time Radiance Field RenderingBernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, George DrettakisSIGGRAPH 2023 · 5,687 次引用

相关 Paper

  1. Taming Mode Collapse in Score Distillation for Text-to-3D GenerationPeihao Wang, Dejia Xu, Zhiwen Fan, Dilin Wang 等CVPR 2024 · 8 次引用
  2. HandDreamer: Zero-Shot Text to 3D Hand Model Generation using Corrective Hand Shape GuidanceGreen Rosh, Prateek Kukreja, Vishakha SR, Pawan Prasad B HCVPR 2026
  3. Debiasing Scores and Prompts of 2D Diffusion for View-consistent Text-to-3D GenerationSusung Hong, Donghoon Ahn, Seungryong KimNeurIPS 2023 · 46 次引用
  4. PlacidDreamer: Advancing Harmony in Text-to-3D GenerationShuo Huang, Shikun Sun, Zixuan Wang, Xiaoyu Qin 等ACM MM 2024 · 4 次引用
  5. DreamControl: Control-Based Text-to-3D Generation with 3D Self-PriorTianyu Huang, Yihan Zeng, Zhilu Zhang, Wan Xu 等CVPR 2024
黄昏的海面,两侧是细线勾勒的悬崖