Research map / Video / Video segmentation and tracking
SIMONe: View-Invariant, Temporally-Abstracted Object Representations via Unsupervised Video Decomposition
Rishabh Kabra, Daniel Zoran, Goker Erdogan, Loic Matthey, Antonia Creswell, Matthew Botvinick, Alexander Lerchner, Christopher P. Burgess
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 SIMONe's joint video decomposition rather than the strongest listed video baseline, MONet | Objects Room 9 video segmentation · Objects Room 9 | ARI-F: 0.865 -> 0.936 | Improved |
| Use SIMONe's joint video decomposition rather than the strongest listed video baseline, S-IODINE | CATER video segmentation · CATER | ARI-F: 0.668 -> 0.918 | Improved |
| Use SIMONe's joint video decomposition rather than the strongest listed video baseline, MONet | Playroom video segmentation · Playroom | ARI-F: 0.442 -> 0.800 | Improved |
| Predict allocentric position using object latent plus timestep instead of object latent alone (ablation) | CATER trajectory probe, all objects · CATER | R²: 0.710 -> 0.871 | Improved |
| Add frame latent to object latent and timestep for trajectory prediction (ablation) | CATER trajectory probe, all objects · CATER | R²: 0.871 -> 0.876 | No clear effect |
| Use the correct object's latent rather than the other object latents to predict its position (ablation) | CATER trajectory probe, all objects · CATER | R²: -0.062 -> 0.871 | Improved |
| Predict allocentric position using object latent plus timestep instead of object latent alone (ablation) | CATER trajectory probe, moving objects · CATER | R²: 0.724 -> 0.894 | Improved |
| Add frame latent to object latent and timestep for trajectory prediction (ablation) | CATER trajectory probe, moving objects · CATER | R²: 0.894 -> 0.898 | No clear effect |
| Use the correct object's latent rather than the other object latents to predict its position (ablation) | CATER trajectory probe, moving objects · CATER | R²: -0.022 -> 0.894 | Improved |
| Use frame latent rather than object latents to predict camera location | CATER camera-pose probe · CATER | R²: 0.044 -> 0.949 | Improved |
| Use frame latent rather than object latents to predict camera orientation | CATER camera-pose probe · CATER | R²: 0.292 -> 0.946 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MONet — MoNet: Deep Motion Exploitation for Video Object Segmentation (CVPR 2018) · better on 3 of 3 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
- Unsupervised Object-Centric Learning From Multiple Unspecified Viewpoints (TPAMI 2024 · similarity 0.89)
- Unsupervised Learning of Compositional Scene Representations from Multiple Unspecified Viewpoints (AAAI 2022 · similarity 0.89)
- OSN: Infinite Representations of Dynamic 3D Scenes from Monocular Videos (ICML 2024 · similarity 0.87)
- Temporally Consistent Object-Centric Learning by Contrasting Slots (CVPR 2025 · similarity 0.87)
- DyST: Towards Dynamic Neural Scene Representations on Real-World Videos (ICLR 2024 · similarity 0.87)
- Reconstruct, Inpaint, Test-Time Finetune: Dynamic Novel-view Synthesis from Monocular Videos (NeurIPS 2025 · similarity 0.86)
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.
