Research map / 3D vision / 3D generation
Correspondence-Attention Alignment for Multi-View Diffusion Models
Minkyung Kwon, Jinhyeok Choi, Jiho Park, Seonghu Jeon, Jinhyuk Jang, Junyoung Seo, Minseop Kwak, Jin-Hwa Kim, Seungryong Kim
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 |
|---|---|---|---|
| supervise one cross-view attention layer (layer 10) with geometric correspondence | CAT3D · RealEstate10K, 80k iters | PSNR: 18.99 -> 19.40 | Improved |
| add CAMEO supervision | CAT3D · DTU out-of-domain, 400k iters | PSNR: 11.16 -> 12.58 | Improved |
| add CAMEO supervision | MVGenMaster · RealEstate10K, 20k iters | PSNR: 17.35 -> 18.31 | Improved |
| add CAMEO supervision | Hunyuan-DiT multi-view model · RealEstate10K, 20k iters | PSNR: 14.40 -> 16.17 | Improved |
| add REPA feature alignment | CAT3D · RealEstate10K, 80k iters | PSNR: 18.99 -> 18.70 | Hurt |
| add Geometry Forcing feature alignment | CAT3D · CO3D, 10k iters | PSNR: 14.55 -> 14.11 | Hurt |
| add MLP projection head on attention logits (ablation) | CAMEO · RealEstate10K, 40k iters | PSNR: 18.08 -> 18.31 | Improved |
| use cross-entropy loss instead of L1 (ablation) | CAMEO · RealEstate10K, 40k iters | PSNR: 17.84 -> 18.31 | Improved |
| use cycle-consistency threshold 1.5 instead of none (ablation) | CAMEO · RealEstate10K, 40k iters | PSNR: 18.18 -> 18.31 | Improved |
| raise loss weight from 0.02 to 0.03 (ablation) | CAMEO · RealEstate10K, 40k iters | LPIPS: 0.337 -> 0.377 | Hurt |
| supervise layer 10 instead of layer 6 (ablation) | CAMEO · RealEstate10K, 40k iters | PSNR: 17.84 -> 18.31 | Improved |
| supervise layer 10 instead of layer 2 (ablation) | CAMEO · RealEstate10K, 80k iters | PSNR: 18.80 -> 19.08 | Improved |
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.
- CAT3D — CAT3D: Create Anything in 3D with Multi-View Diffusion Models (NeurIPS 2024) · better on 8 of 8 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
- SweetDreamer: Aligning Geometric Priors in 2D diffusion for Consistent Text-to-3D (ICLR 2024 · similarity 0.90)
- Understanding Multi-View Transformers (NeurIPS 2026 · similarity 0.88)
- MVGenMaster: Scaling Multi-View Generation from Any Image via 3D Priors Enhanced Diffusion Model (CVPR 2025 · similarity 0.88)
- EpiDiff: Enhancing Multi-View Synthesis via Localized Epipolar-Constrained Diffusion (CVPR 2024 · similarity 0.88)
- Carve3D: Improving Multi-view Reconstruction Consistency for Diffusion Models with RL Finetuning (CVPR 2024 · similarity 0.88)
- Unleashing Diffusion Transformers for Visual Correspondence by Modulating Massive Activations (NeurIPS 2025 · 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.
