Research map / 3D vision / 3D generation
RUST: Latent Neural Scene Representations From Unposed Imagery
Mehdi S. M. Sajjadi, Aravindh Mahendran, Thomas Kipf, Etienne Pot, Daniel Duckworth, Mario Lučić, Klaus Greff
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 |
|---|---|---|---|
| Train and infer without camera poses, using a target-image-derived latent pose | RUST · MSN | PSNR: 18.65 -> 23.49 | Improved |
| Train and infer without camera poses, using a target-image-derived latent pose | RUST · Street View | PSNR: 21.25 -> 22.50; 23.63 -> 22.50 | Mixed |
| Use the right rather than the left target-image half for pose estimation (ablation) | RUST · MSN | PSNR: 23.49 -> 23.88 (+0.39 db) | Improved |
| Stop gradients from the latent pose into the encoder (ablation) | RUST · MSN | PSNR: 23.49 -> 23.16 (drop of 0.33 db) | Hurt |
| Remove Pose Estimator cross-attention into the SLSR (ablation) | RUST · MSN | PSNR: 23.49 -> 20.83 (drop of 2.66 db) | Hurt |
| Remove self-attention between target-image tokens (ablation) | RUST · MSN | PSNR: 23.49 -> 22.97 (drop of 0.52 db) | Hurt |
| Reduce latent pose size to 3 dimensions (ablation) | RUST · MSN | PSNR: 23.49 -> 20.40 (-3.09 db) | Hurt |
| Increase latent pose size to 64 dimensions (ablation) | RUST · MSN | PSNR: 23.49 -> 23.40 | No clear effect |
| Increase latent pose size to 768 dimensions (ablation) | RUST · MSN | PSNR: 23.49 -> 23.11 | No clear effect |
Later papers that compared with this method
3 later accepted papers report RUST as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- True Self-Supervised Novel View Synthesis is Transferable (ICLR 2026 · better on 8 of 8 shared results)
- ERUPT: Efficient Rendering with Unposed Patch Transformer (CVPR 2025 · better on 3 of 3 shared results)
- FlowCam: Training Generalizable 3D Radiance Fields without Camera Poses via Pixel-Aligned Scene Flow (NeurIPS 2023 · better on 8 of 8 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
- Scene Representation Transformer: Geometry-Free Novel View Synthesis Through Set-Latent Scene Representations (CVPR 2022 · similarity 0.89)
- UMAMI: Unifying Masked Autoregressive Models and Deterministic Rendering for View Synthesis (NeurIPS 2025 · similarity 0.87)
- NViST: In the Wild New View Synthesis from a Single Image with Transformers (CVPR 2024 · similarity 0.86)
- LVSM: A Large View Synthesis Model with Minimal 3D Inductive Bias (ICLR 2025 · similarity 0.86)
- DyST: Towards Dynamic Neural Scene Representations on Real-World Videos (ICLR 2024 · similarity 0.86)
- ERUPT: Efficient Rendering with Unposed Patch Transformer (CVPR 2025 · similarity 0.85)
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.
