Research map / 3D vision / Gaussian splatting
Scaling Transformer-Based Novel View Synthesis with Models Token Disentanglement and Synthetic Data
Nithin Gopalakrishnan Nair, Srinivas Kaza, Xuan Luo, Vishal M. Patel, Stephen Lombardi, Jungyeon Park
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 shared LVSM blocks with Tok-D source/target modulation | LVSM · RealEstate10k | PSNR: 28.89 -> 30.02 | Improved |
| Tok-D versus LVSM | LVSM · DL3DV in-distribution | PSNR: 19.91 -> 21.55 | Improved |
| Tok-D versus LVSM | LVSM · ACID in-distribution | PSNR: 29.19 -> 29.47 | Improved |
| add synthetic data to training (ablation) | LVSM · RealEstate10k | PSNR: 28.89 -> 28.49 | Hurt |
| add synthetic data to training (ablation) | LVSM · DL3DV cross-dataset | PSNR: 20.52 -> 20.21 | Hurt |
| add synthetic data to training (ablation) | Tok-D transformer · DL3DV cross-dataset | PSNR: 21.18 -> 21.27 | Improved |
| add synthetic data to training (ablation) | Tok-D transformer · RealEstate10k | PSNR: 30.02 -> 29.97 | No clear effect |
| add real DL3DV data to training (ablation) | Tok-D transformer · DL3DV | PSNR: 21.27 -> 23.14 | Improved |
| add real DL3DV data to training (ablation) | LVSM · DL3DV | PSNR: 20.21 -> 21.37 | Improved |
| add pre-modulation of whole block input (ablation) | Tok-D transformer · RealEstate10k | PSNR: 28.50 -> 29.69 | Improved |
| add separate pre-modulation in attention and MLP only (ablation) | Tok-D transformer · RealEstate10k | PSNR: 28.50 -> 28.51 | No clear effect |
| add post-modulation scaling on top of pre-modulation (ablation) | Tok-D transformer · RealEstate10k | PSNR: 28.51 -> 30.02 | Improved |
Closest papers by meaning
- LVSM: A Large View Synthesis Model with Minimal 3D Inductive Bias (ICLR 2025 · similarity 0.90)
- Novel View Synthesis with Pixel-Space Diffusion Models (CVPR 2025 · similarity 0.89)
- NViST: In the Wild New View Synthesis from a Single Image with Transformers (CVPR 2024 · similarity 0.89)
- Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models (CVPR 2025 · similarity 0.88)
- UMAMI: Unifying Masked Autoregressive Models and Deterministic Rendering for View Synthesis (NeurIPS 2025 · similarity 0.88)
- Déjà View: Looping Transformers for Multi-View 3D Reconstruction (NeurIPS 2026 · 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.
