Research map / 3D vision / 3D generation
LagerNVS: Latent Geometry for Fully Neural Real-time Novel View Synthesis
Stanislaw Szymanowicz, Minghao Chen, Jianyuan Wang, Christian Rupprecht, Andrea Vedaldi
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 |
|---|---|---|---|
| initialize encoder from VGGT (3D-aware) instead of training from scratch (ablation) | LagerNVS highway encoder-decoder · DL3DV 2-view, known cameras | PSNR: 18.03 -> 21.02 | Improved |
| initialize from DinoV2 (2D pre-training) instead of scratch (ablation) | LagerNVS highway encoder-decoder · DL3DV 2-view, known cameras | PSNR: 18.03 -> 18.17 | Improved |
| fine-tune the whole VGGT encoder end-to-end instead of freezing it (ablation) | LagerNVS · DL3DV 2-view, known cameras | PSNR: 19.01 -> 21.02 | Improved |
| use highway encoder-decoder instead of decoder-only (ablation) | NVS architecture · DL3DV 2-view, known cameras | PSNR: 18.39 -> 21.02 | Improved |
| use highway encoder-decoder instead of bottleneck encoder-decoder (ablation) | NVS architecture · DL3DV 2-view, known cameras | PSNR: 17.53 -> 21.02 | Improved |
| use full attention decoder instead of cross-attention decoder (ablation) | LagerNVS · RealEstate10k 2-view (batch 512) | PSNR: 31.06 -> 31.39 | Improved |
| use cross-attention decoder instead of full attention (speed trade-off) (ablation) | LagerNVS · DL3DV 2-view, known cameras | PSNR: 21.02 -> 21.30 | Mixed |
| replace LVSM with LagerNVS highway encoder-decoder | RealEstate10k 2-view, batch 512 · RealEstate10k | PSNR: 29.67 -> 31.39 | Improved |
| replace explicit 3DGS (DepthSplat) with latent 3D representation | DL3DV 4-view with cameras · DL3DV 4-view | PSNR: 22.30 -> 27.56 | Improved |
| replace AnySplat with LagerNVS on CO3D without cameras | CO3D 9-view · CO3D 9-view | PSNR: 15.87 -> 22.37 | Improved |
| fine-tune decoder with diffusion objective (frozen encoder) | LagerNVS · occlusion and extrapolation (qualitative) | visual quality | Improved |
Later papers that compared with this method
1 later accepted paper reports LagerNVS as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Less Decoder is More Encoder: Geometric Representation Learning from Novel View Synthesis (NeurIPS 2026 · better on 3 of 4 shared results)
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.
- DepthSplat — DepthSplat: Connecting Gaussian Splatting and Depth (CVPR 2025) · better on 2 of 2 shared results
- Flare — FLARE: Feed-forward Geometry, Appearance and Camera Estimation from Uncalibrated Sparse Views (CVPR 2025) · 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
- LVSM: A Large View Synthesis Model with Minimal 3D Inductive Bias (ICLR 2025 · similarity 0.89)
- WildFusion: Learning 3D-Aware Latent Diffusion Models in View Space (ICLR 2024 · similarity 0.88)
- Less Decoder is More Encoder: Geometric Representation Learning from Novel View Synthesis (NeurIPS 2026 · similarity 0.87)
- UMAMI: Unifying Masked Autoregressive Models and Deterministic Rendering for View Synthesis (NeurIPS 2025 · similarity 0.87)
- EscherNet: A Generative Model for Scalable View Synthesis (CVPR 2024 · similarity 0.87)
- Efficient-3Dim: Learning a Generalizable Single-image Novel-view Synthesizer in One Day (ICLR 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.
