Research map / 3D vision / 3D generation
DreamComposer++: Empowering Diffusion Models with Multi-View Conditions for 3D Content Generation
Yunhan Yang, Shuo Chen, Yukun Huang, Xiaoyang Wu, Yuan-Chen Guo, Edmund Y. M. Lam, Hengshuang Zhao, Tong He, Xihui Liu
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 |
|---|---|---|---|
| add DreamComposer++ multi-view conditioning | Zero-1-to-3 · GSO NVS, 4 views, elevation 0 | PSNR: 20.82 -> 25.25 | Improved |
| add DreamComposer++ multi-view conditioning | Zero-1-to-3 · GSO NVS, 4 views, elevation 30 | LPIPS: 0.128 -> 0.086 | Improved |
| add DreamComposer++ with generated extra view | SyncDreamer · GSO NVS | PSNR: 20.05 -> 20.52 | Improved |
| add DreamComposer++ with 2 views | SV3D · GSO NVS, elevation 0 | LPIPS: 0.3156 -> 0.1441 | Improved |
| increase input views from 2 to 6 | DC-SV3D · GSO NVS, elevation 0 | PSNR: 19.83 -> 22.21 | Improved |
| increase input views from 2 to 6 | DC-Zero-1-to-3 · GSO NVS, elevation 0 | PSNR: 20.38 -> 26.99 | Improved |
| freeze the pretrained diffusion UNet (ablation) | DC-SyncDreamer · GSO NVS | PSNR: 15.96 -> 20.52 | Improved |
| add reconstruction MSE loss in stage two (ablation) | DC-SyncDreamer · GSO NVS | PSNR: 16.18 -> 20.52 | Improved |
| add view-conditioning cross-attention in 3D lifting (ablation) | DC-SyncDreamer · GSO NVS | PSNR: 19.04 -> 20.52 | Improved |
| add view-aware attention (ablation) | DC-SV3D (2 views) · GSO NVS, elevation 0 | LPIPS: 0.2968 -> 0.1441 | Improved |
| use attention-based fusion instead of convolution (ablation) | DC-SV3D (2 views) · GSO NVS, elevation 0 | PSNR: 17.98 -> 19.83 | Improved |
| use ground-truth latents instead of Lift3D rendered features (ablation) | DC-SV3D · video generation | visual quality | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- Zero-1-to-3 — Zero-1-to-3: Zero-shot One Image to 3D Object (ICCV 2023) · better on 7 of 7 shared results
- ViewFormer — ViewFormer: NeRF-Free Neural Rendering from Few Images Using Transformers (ECCV 2022) · better on 3 of 3 shared results
- SyncDreamer — SyncDreamer: Generating Multiview-consistent Images from a Single-view Image (ICLR 2024) · better on 2 of 2 shared results
- RealFusion — RealFusion: 360deg Reconstruction of Any Object From a Single Image (CVPR 2023) · better on 1 of 1 shared result
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
- DreamComposer: Controllable 3D Object Generation via Multi-View Conditions (CVPR 2024 · similarity 0.97)
- MVDream: Multi-view Diffusion for 3D Generation (ICLR 2024 · similarity 0.90)
- DreamCraft3D: Hierarchical 3D Generation with Bootstrapped Diffusion Prior (ICLR 2024 · similarity 0.89)
- ViVid-1-to-3: Novel View Synthesis with Video Diffusion Models (CVPR 2024 · similarity 0.89)
- Sharp-It: A Multi-view to Multi-view Diffusion Model for 3D Synthesis and Manipulation (CVPR 2025 · similarity 0.89)
- Diffusion$^2$: Dynamic 3D Content Generation via Score Composition of Video and Multi-view Diffusion Models (ICLR 2025 · 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.
