Research map / 3D vision / 3D generation
Efficient View Synthesis and 3D-Based Multi-Frame Denoising With Multiplane Feature Representations
Thomas Tanay, Aleš Leonardis, Matteo Maggioni
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 |
|---|---|---|---|
| Replace fixed MPI overcompositing with the MPFER learnable renderer and feature representation | MPI view synthesis · Spaces, 12 input views | PSNR: 30.85 -> 35.73 | Improved |
| Use the MPFER encoder-renderer instead of 2D burst denoising | Multi-frame denoising · Spaces, Gain 20 | PSNR: 31.32 -> 33.23 | Improved |
| Use clean-image camera poses instead of poses computed on noisy images (ablation) | MPFER · LLFF-N synthetic denoising, Gain 20 | PSNR: 26.23 -> 27.71 | Improved |
| Apply MPFER rather than NAN to novel-view synthesis under noisy conditions | Noisy-input view synthesis · LLFF-N, Gain 20 | PSNR: 22.37 -> 23.72 | Improved |
| Use one-shot concatenation across views and depths rather than the MPFER design (ablation) | MPI view synthesis · Spaces, 12 input views | PSNR: 35.73 -> 27.43 | Hurt |
| Process each depth independently without cross-depth communication (ablation) | MPI view synthesis · Spaces, 12 input views | PSNR: 35.73 -> 30.70 | Hurt |
| Add a second depthwise refinement pass with gradient components (ablation) | MPINet-dw · Spaces, 12 input views | PSNR: 30.70 -> 30.85 | Improved |
| Increase depth planes from 16 to 64 (ablation) | MPFER · Spaces denoising, Gain 20 | PSNR: 33.00 -> 33.23 | Improved |
| Increase feature channels from 4 to 16 (ablation) | MPFER · Spaces denoising, Gain 20 | PSNR: 32.74 -> 33.14 | Improved |
| Increase the PSV/MPF upscaling factor from 1.0 to 1.5 (ablation) | MPFER · Spaces denoising, Gain 20 | PSNR: 32.89 -> 33.08 | 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.
- DeepView — DeepView: View Synthesis With Learned Gradient Descent (CVPR 2019) · better on 4 of 4 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
- DMV3D: Denoising Multi-view Diffusion Using 3D Large Reconstruction Model (ICLR 2024 · similarity 0.88)
- SIR-DIFF: Sparse Image Sets Restoration with Multi-View Diffusion Model (CVPR 2025 · similarity 0.88)
- Denoising Diffusion via Image-Based Rendering (ICLR 2024 · similarity 0.87)
- ViewFusion: Towards Multi-View Consistency via Interpolated Denoising (CVPR 2024 · similarity 0.87)
- NAMVIS: Next-Scale Autoregressive Multi-View Image Synthesis (NeurIPS 2026 · similarity 0.87)
- LVSM: A Large View Synthesis Model with Minimal 3D Inductive Bias (ICLR 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.
