Research map / 3D vision / 3D generation
From Rays to Projections: Better Inputs for Feed-Forward View Synthesis
Zirui Wu, Zeren Jiang, Martin R. Oswald, Jie Song
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 projective conditioning (point cloud projection image) to the LVSM baseline (ablation) | LVSM 12 layers · RealEstate10K (Total) | PSNR: 24.60 -> 25.20 | Improved |
| add DINOv3 feature prior on top of projection (ablation) | PVSM without pretraining · RealEstate10K (Total) | PSNR / LPIPS: PSNR 25.20 -> 25.13, LPIPS 0.177 -> 0.163 | Mixed |
| add MAE pretraining on DL3DV (ablation) | PVSM with DINO and projection · RealEstate10K (Total) | PSNR: 25.13 -> 25.64 | Improved |
| pretrain on large DL3DV instead of RealEstate10K (ablation) | PVSM 12 layers · RealEstate10K (Total) | PSNR: 24.78 -> 25.64 | Improved |
| increase finetuning from 200 steps to 50K steps after DL3DV pretraining (ablation) | PVSM 12 layers · RealEstate10K (Total) | PSNR: 22.32 -> 25.64 | Improved |
| add RoPE to the projective model (ablation) | PVSM · Single-object overfit | PSNR: 21.18 -> 30.03 | Improved |
| add RoPE to LVSM (ablation) | LVSM · Single-object overfit | PSNR: 25.39 -> 25.88 | Improved |
| replace Plucker ray conditioning with projective conditioning | LVSM · Consistency Benchmark (World Scale) | PSNR (M): 14.56 -> 25.43 | Improved |
| fine-tune with 500 steps of camera augmentation | LVSM · Consistency Benchmark (World Scale) | PSNR (M): 14.56 -> 13.57 | Hurt |
| fine-tune with 500 steps of camera augmentation | PVSM · Consistency Benchmark (World Scale) | PSNR (M): 25.43 -> 25.78 | Improved |
| projective conditioning vs Plucker rays under roll | PVSM vs LVSM · Consistency Benchmark (Roll) | PSNR (M): 19.54 -> 17.53 | Hurt |
| scale model from 12 to 24 layers | PVSM · RealEstate10K (Total) | PSNR: 25.64 -> 26.90 | Improved |
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.
- RayZer — RayZer: A Self-supervised Large View Synthesis Model (ICCV 2025) · better on 1 of 1 shared result
- PixelSplat — pixelSplat: 3D Gaussian Splats from Image Pairs for Scalable Generalizable 3D Reconstruction (CVPR 2024) · better on 1 of 1 shared result
- MVSplat — MVSplat: Efficient 3D Gaussian Splatting from Sparse Multi-View Images (ECCV 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
- NAMVIS: Next-Scale Autoregressive Multi-View Image Synthesis (NeurIPS 2026 · similarity 0.88)
- UMAMI: Unifying Masked Autoregressive Models and Deterministic Rendering for View Synthesis (NeurIPS 2025 · similarity 0.88)
- Novel View Synthesis from A Few Glimpses via Test-Time Natural Video Completion (NeurIPS 2025 · similarity 0.87)
- Less Decoder is More Encoder: Geometric Representation Learning from Novel View Synthesis (NeurIPS 2026 · similarity 0.87)
- Cameras as Relative Positional Encoding (NeurIPS 2025 · similarity 0.87)
- CRAYM: Neural Field Optimization via Camera RAY Matching (NeurIPS 2024 · 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.
