Research map / 3D vision / 3D generation
MV-Adapter: Multi-View Consistent Image Generation Made Easy
Zehuan Huang, Yuan-Chen Guo, Haoran Wang, Ran Yi, Lizhuang Ma, Yan-Pei Cao, Lu Sheng
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 parallel instead of serial attention architecture (ablation) | MV-Adapter (SDXL) · GSO image-to-multiview | PSNR: 20.687 -> 22.131 | Improved |
| use parallel instead of serial attention architecture (ablation) | MV-Adapter (SDXL) · GSO image-to-multiview | SSIM: 0.8681 -> 0.8816 | Improved |
| use parallel instead of serial attention architecture (ablation) | MV-Adapter (SDXL) · GSO image-to-multiview | LPIPS: 0.1149 -> 0.1002 | Improved |
| scale base model from SD2.1 to SDXL | MV-Adapter · Objaverse text-to-multiview | FID: 31.24 -> 29.71 | Improved |
| scale base model from SD2.1 to SDXL | MV-Adapter · GSO image-to-multiview | PSNR: 20.867 -> 22.131 | Improved |
| scale base model from SD2.1 to SDXL | MV-Adapter · 3D reconstruction | Chamfer Distance: 0.0317 -> 0.0206 | Improved |
| condition on image instead of text for texture generation | MV-Adapter (SDXL) · 3D texture generation | FID: 32.75 -> 27.28 | Improved |
| adapter training instead of full fine-tuning (vs Era3D) | MV-Adapter (SD2.1) · training cost | trainable parameters: 993M -> 127M | Improved |
| adapter training instead of full fine-tuning (vs Era3D) | MV-Adapter (SDXL) · training cost | trainable parameters: 3.1B -> 490M | Improved |
| adapter training instead of full fine-tuning (vs Era3D) | MV-Adapter (SD2.1) · training speed | iter/s: 2.2iter/s -> 3.1iter/s | Improved |
| decoupled adapter instead of full-tuned MVDream with LoRA (ablation) | MV-Adapter · community T2I derivatives | alignment with text and style | Improved |
| multi-view LoRA instead of decoupled attention (ablation) | multi-view LoRA · switching base model or adding LoRA | multi-view consistency | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- Ouroboros3D — Ouroboros3D: Image-to-3D Generation via 3D-aware Recursive Diffusion (CVPR 2025) · better on 2 of 2 shared results
- SPAD — SPAD: Spatially Aware Multi-View Diffusers (CVPR 2024) · better on 2 of 2 shared results
- Paint3D — Paint3D: Paint Anything 3D with Lighting-Less Texture Diffusion Models (CVPR 2024) · 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
- MVGenMaster: Scaling Multi-View Generation from Any Image via 3D Priors Enhanced Diffusion Model (CVPR 2025 · similarity 0.90)
- MVDream: Multi-view Diffusion for 3D Generation (ICLR 2024 · similarity 0.88)
- IT3D: Improved Text-to-3D Generation with Explicit View Synthesis (AAAI 2024 · similarity 0.88)
- ViewDiff: 3D-Consistent Image Generation with Text-to-Image Models (CVPR 2024 · similarity 0.88)
- MVBoost: Boost 3D Reconstruction with Multi-View Refinement (CVPR 2025 · similarity 0.88)
- Vivid-ZOO: Multi-View Video Generation with Diffusion Model (NeurIPS 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.
