Research map / Video / Video segmentation and tracking
SlotFormer: Unsupervised Visual Dynamics Simulation with Object-Centric Models
Ziyi Wu, Nikita Dvornik, Klaus Greff, Thomas Kipf, Animesh Garg
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 |
|---|---|---|---|
| Image reconstruction loss L_I added to slot dynamics training (ablation) | SlotFormer (SAVi-based) · OBJ3D | LPIPS: 0.093 -> 0.080 | Improved |
| Teacher forcing instead of rollout-based training with predicted slots (ablation) | SlotFormer · OBJ3D | LPIPS: 0.080 -> 0.106 | Hurt |
| Teacher forcing instead of rollout-based training (ablation) | SlotFormer · Physion | Improved accuracy: 1.9 -> 0.2 | Hurt |
| Temporal-only positional encoding instead of naive per-slot sinusoidal encoding (ablation) | SlotFormer · Physion | Improved accuracy: 1.6 -> 1.9 | Improved |
| Temporal-only positional encoding instead of naive per-slot sinusoidal encoding (ablation) | SlotFormer · OBJ3D | LPIPS: 0.082 vs 0.080 | No clear effect |
| Burn-in length T=3 instead of T=6 default (ablation) | SlotFormer · OBJ3D | PSNR: 32.40 -> 31.26 | Hurt |
| Transformer layers increased to 8 (from 4) (ablation) | SlotFormer · OBJ3D | LPIPS: 0.080 -> 0.087 | Hurt |
| Transformer layers increased to 4 (from 8) on Physion (ablation) | SlotFormer · Physion | Improved accuracy: 1.9 -> 1.3 | Hurt |
| Autoregressive SlotFormer rollout with SlotFormer predicted future slots fed to Aloe | Aloe on CLEVRER predictive questions · CLEVRER predictive VQA | per question accuracy: 87.3 -> 93.3 | Improved |
| Unsupervised SlotFormer dynamics rollout added to planning | SAVi success classifier on PHYRE · PHYRE-1B within-template | AUCCESS: 80.7 -> 82.0 | Improved |
| Object-centric slot dynamics instead of pixel-based PredRNN | video prediction · OBJ3D | LPIPS: 0.12 -> 0.08 | Improved |
| SlotFormer dynamics on slots vs VQ patch tokens (VQFormer) | video prediction · CLEVRER | LPIPS: 0.18 -> 0.11 | Improved |
Later papers that compared with this method
3 later accepted papers report SlotFormer as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Factored Latent Action World Models (ICML 2026 · better on 4 of 4 shared results)
- OCK: Unsupervised Dynamic Video Prediction with Object-Centric Kinematics (ICCV 2025 · better on 8 of 8 shared results)
- VDT: General-purpose Video Diffusion Transformers via Mask Modeling (ICLR 2024 · better on 0 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- PredRNN — PredRNN: Recurrent Neural Networks for Predictive Learning using Spatiotemporal LSTMs (NeurIPS 2017) · better on 2 of 5 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
- Parallelized Spatiotemporal Slot Binding for Videos (ICML 2024 · similarity 0.88)
- Slot-VLM: Object-Event Slots for Video-Language Modeling (NeurIPS 2024 · similarity 0.87)
- Slot State Space Models (NeurIPS 2024 · similarity 0.87)
- Temporally Consistent Object-Centric Learning by Contrasting Slots (CVPR 2025 · similarity 0.87)
- PlaySlot: Learning Inverse Latent Dynamics for Controllable Object-Centric Video Prediction and Planning (ICML 2025 · similarity 0.87)
- Slot-Guided Adaptation of Pre-trained Diffusion Models for Object-Centric Learning and Compositional Generation (ICLR 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.
