Research map / Image generation / Autoregressive and latent generation
SemanticNVS: Improving Semantic Scene Understanding in Generative Novel View Synthesis
Xinya Chen, Christopher Wewer, Jiahao Xie, Xinting Hu, Jan Eric Lenssen
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 warped DINO features to warped-RGB baseline (ablation) | SEVA + Warped RGB · RealEstate10K long | FID: 29.942 -> 28.908 | Improved |
| add iterative DINO features of the denoised estimate (ablation) | SEVA + Warped DINO · RealEstate10K long | FID: 28.908 -> 26.561 | Improved |
| condition on RGB of the x0 estimate instead of its DINO features (ablation) | SemanticNVS · RealEstate10K long | FID: 26.561 -> 28.260 | Hurt |
| add warped RGB conditioning (ablation) | SEVA finetuned · RealEstate10K long | FID: 31.896 -> 29.942 | Improved |
| replace DINOv2 features with DINOv3 (ablation) | SemanticNVS · RealEstate10K long | FID: 26.561 -> 29.058 | Hurt |
| replace DINOv2 features with VGGT features (ablation) | SemanticNVS · RealEstate10K long | FID: 26.561 -> 28.704 | Hurt |
| use REPA feature alignment instead of explicit semantic conditioning (ablation) | SEVA finetuned · RealEstate10K long | FID: 26.561 -> 30.579 | Hurt |
| use smaller DINOv2 encoder (ViT-S vs ViT-L) (ablation) | SemanticNVS · RealEstate10K long | FID: 26.561 -> 28.483 | Hurt |
| use SemanticNVS only for anchor views and SEVA interpolation (ablation) | SemanticNVS · RealEstate10K full trajectory | FID: 21.84 -> 21.90 | No clear effect |
| add semantic conditioning (SemanticNVS) vs SEVA | SEVA · RealEstate10K full trajectory | FID: 25.347 -> 21.842 | Improved |
| add iterative DINO step-wise conditioning (ablation) | SemanticNVS · RealEstate10K long | inference time (s/img): 1.303 -> 1.404 | Hurt |
Closest papers by meaning
- Novel View Synthesis with Pixel-Space Diffusion Models (CVPR 2025 · similarity 0.90)
- NVComposer: Boosting Generative Novel View Synthesis with Multiple Sparse and Unposed Images (CVPR 2025 · similarity 0.88)
- TOSS: High-quality Text-guided Novel View Synthesis from a Single Image (ICLR 2024 · similarity 0.87)
- MVGenMaster: Scaling Multi-View Generation from Any Image via 3D Priors Enhanced Diffusion Model (CVPR 2025 · similarity 0.87)
- LVSM: A Large View Synthesis Model with Minimal 3D Inductive Bias (ICLR 2025 · similarity 0.87)
- NAMVIS: Next-Scale Autoregressive Multi-View Image Synthesis (NeurIPS 2026 · 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.
