Research map / 3D vision / 3D generation
UniVerse: Unleashing the Scene Prior of Video Diffusion Models for Robust Radiance Field Reconstruction
Jin Cao, Hongrui Wu, Ziyong Feng, Hujun Bao, Xiaowei Zhou, Sida Peng
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 |
|---|---|---|---|
| sort images with ThreadPose when forming the initial video (ablation) | UniVerse · Synthetic (image-to-video ablation) | PSNR: 18.25 -> 20.71 | Improved |
| insert zero frames (implicit views) into ordered images (ablation) | UniVerse · Synthetic (image-to-video ablation) | PSNR: 17.29 -> 20.71 | Improved |
| sort and insert zero frames instead of directly stacking (ablation) | UniVerse · Synthetic (image-to-video ablation) | SSIM: 0.4598 -> 0.7708 | Improved |
| replace single-input Query Transformer with Multi-input Query Transformer (ablation) | UniVerse · Query Transformer ablation | PSNR: 16.80 -> 17.42 | Improved |
| replace single-input Query Transformer with Multi-input Query Transformer (ablation) | UniVerse · Query Transformer ablation | LPIPS: 0.5015 -> 0.4549 | Improved |
| replace MSE with consistency loss for VDM training (ablation) | UniVerse · Loss ablation | PSNR: 16.19 -> 17.42 | Improved |
| replace MSE with consistency loss for VDM training (ablation) | UniVerse · Loss ablation | LPIPS: 0.5340 -> 0.4549 | Improved |
| add inpainting masks to the VDM input (ablation) | UniVerse · NeRF-on-the-go subset (occlusions only) | visual inpainting | Improved |
| add style mask to the VDM input (ablation) | UniVerse · visual | tone control | Improved |
| restore images with UniVerse before reconstruction instead of ZipNeRF w/GLO | ZipNeRF w/GLO · Synthetic LLFF (sRGB) | PSNR: 14.53 -> 18.09 | Improved |
| restore images with UniVerse before reconstruction instead of ZipNeRF w/GLO | ZipNeRF w/GLO · Real-world captured (sRGB) | PSNR: 16.14 -> 19.65 | Improved |
| add GLO appearance embeddings to ZipNeRF | ZipNeRF · Synthetic LLFF (sRGB) | PSNR: 11.52 -> 14.53 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- WildGaussians — WildGaussians: 3D Gaussian Splatting In the Wild (NeurIPS 2024) · better on 2 of 2 shared results
- ZipNeRF — Zip-NeRF: Anti-Aliased Grid-Based Neural Radiance Fields (ICCV 2023) · better on 2 of 2 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
- Sparse3D: Distilling Multiview-Consistent Diffusion for Object Reconstruction from Sparse Views (AAAI 2024 · similarity 0.88)
- Denoising Diffusion via Image-Based Rendering (ICLR 2024 · similarity 0.87)
- Robust Inverse Graphics via Probabilistic Inference (ICML 2024 · similarity 0.87)
- Coherent 3D Scene Diffusion From a Single RGB Image (NeurIPS 2024 · similarity 0.87)
- Decompositional Neural Scene Reconstruction with Generative Diffusion Prior (CVPR 2025 · similarity 0.87)
- Wonderland: Navigating 3D Scenes from a Single Image (CVPR 2025 · similarity 0.87)
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.
