Research map / 3D vision / 3D generation
Object Scene Representation Transformer
Mehdi S. M. Sajjadi, Daniel Duckworth, Aravindh Mahendran, Sjoerd van Steenkiste, Filip Pavetic, Mario Lucic, Leonidas Guibas, Klaus Greff, Thomas Kipf
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 |
|---|---|---|---|
| Use OSRT instead of ObSuRF | Novel view synthesis and scene decomposition · MSN-Hard | FG-ARI: 0.280 -> 0.619 | Improved |
| Use OSRT instead of ObSuRF | Novel view synthesis and scene decomposition · MSN-Easy | PSNR: 27.41 -> 29.74 | Improved |
| Use OSRT instead of ObSuRF | Novel view synthesis and scene decomposition · CLEVR-3D | FG-ARI: 0.978 -> 0.976 | No clear effect |
| Increase input views from 1 to 5 | OSRT · MSN-Hard | FG-ARI: 0.619 -> 0.812 | Improved |
| Use light-field rendering and the Slot Mixer instead of ObSuRF's volumetric rendering | Novel-view rendering · MSN-Hard | FPS: 0.01 -> 32.5 | Improved |
| Replace Slot Mixer with the SRT Decoder (ablation) | OSRT · MSN-Hard | FG-ARI: 0.812 -> 0.330 | Hurt |
| Replace Slot Mixer with the Spatial Broadcast Decoder (ablation) | OSRT · MSN-Hard | FPS: 32.47 -> 1.39 | Hurt |
| Replace novel-view synthesis with reconstruction of the input views (ablation) | OSRT · MSN-Hard | FG-ARI: 0.812 -> 0.198 | Hurt |
| Increase input views from 1 to 5 (ablation) | OSRT · MSN-Hard | PSNR: 20.52 -> 23.54 | Improved |
| Evaluate with grayscale images instead of RGB (ablation) | RGB-trained OSRT · MSN-Hard grayscale test set | FG-ARI: 0.812 -> 0.780 | Hurt |
| Use volumetric rendering (ablation) | OSRT · MSN-Hard | FG-ARI: 0.812 -> 0.767 | Hurt |
| Remove input-view poses (ablation) | OSRT · MSN-Hard | FG-ARI: 0.812 -> 0.798 | Hurt |
Closest papers by meaning
- ROOTS: Object-Centric Representation and Rendering of 3D Scenes (JMLR 2021 · similarity 0.90)
- Scene Representation Transformer: Geometry-Free Novel View Synthesis Through Set-Latent Scene Representations (CVPR 2022 · similarity 0.89)
- Object-X: Learning to Reconstruct Multi-Modal 3D Object Representations (NeurIPS 2025 · similarity 0.89)
- DORSal: Diffusion for Object-centric Representations of Scenes $\textit{et al.}$ (ICLR 2024 · similarity 0.88)
- OSN: Infinite Representations of Dynamic 3D Scenes from Monocular Videos (ICML 2024 · similarity 0.88)
- PartCrafter: Structured 3D Mesh Generation via Compositional Latent Diffusion Transformers (NeurIPS 2025 · 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.
