Research map / 3D vision / 3D generation
Shepherding Slots to Objects: Towards Stable and Robust Object-Centric Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 ARK and IPPE to Slot Attention (ablation) | Slot Attention · CLEVRTEX, PTR, MOVi-C; CLEVR6 on mIoU | mIoU: 22.2 ± 4.3 -> 34.7 ± 5.3; 17.6 ± 14.7 -> 44.1 ± 9.6; 23.0 ± 9.8 -> 27.7 ± 5.9; 49.5 ± 20.5 -> 63.6 ± 4.3 | Helped |
| Add ARK to Slot Attention (ablation) | Slot Attention · CLEVR6, CLEVRTEX, PTR, MOVi-C | mIoU stability: 49.5 ± 20.5 -> 64.1 ± 3.1; 22.2 ± 4.3 -> 31.4 ± 6.6; 17.6 ± 14.7 -> 43.8 ± 3.0; 23.0 ± 9.8 -> 26.2 ± 6.1 | Helped |
| Add IPPE to Slot Attention (ablation) | Slot Attention · MOVi-C | mIoU: 23.0 ± 9.8 -> 27.2 ± 7.9 | Helped |
| Use IPPE without ARK (ablation) | Slot Attention · CLEVR6 and CLEVRTEX | mIoU: 49.5 ± 20.5 -> 57.8 ± 14.6; 22.2 ± 4.3 -> 25.1 ± 7.4 | Helped |
| Increase attention temperature to τ=2 (ablation) | Slot Attention · PTR | mIoU: 17.6 ± 14.7 -> 34.3 ± 12.0 | Helped |
| Apply Gaussian smoothing (ablation) | Slot Attention · CLEVRTEX | mIoU: 22.2 ± 4.3 -> 26.0 ± 8.5 | Helped |
| Replace WNConv with unconstrained Conv (ablation) | ARK · PTR | mIoU: 43.8 ± 3.0 -> 12.4 ± 9.7 | Hurt |
| Use WNConv rather than the tested smoothing alternatives (ablation) | ARK · CLEVR6, CLEVRTEX, PTR, MOVi-C | mIoU: 64.1 ± 3.1; 31.4 ± 6.6; 43.8 ± 3.0; 27.2 ± 6.1 | Helped |
| Use SLASH rather than the strongest listed baseline | Object discovery methods · CLEVRTEX, PTR, MOVi-C | mIoU: 29.2 ± 0.8 -> 34.7 ± 5.3; 28.5 ± 11.3 -> 44.1 ± 9.6; 23.0 ± 9.8 -> 27.7 ± 5.9 | Helped |
| Use SLASH rather than WS-SA | Object discovery methods · CLEVR6 | ARI: 90.7 ± 2.0 -> 90.3 ± 4.3 | No clear effect |
Closest papers by meaning
- Bootstrapping Top-down Information for Self-modulating Slot Attention (NeurIPS 2024 · similarity 0.92)
- MetaSlot: Break Through the Fixed Number of Slots in Object-Centric Learning (NeurIPS 2025 · similarity 0.91)
- Adaptive Slot Attention: Object Discovery with Dynamic Slot Number (CVPR 2024 · similarity 0.90)
- Object-Centric Pretraining via Target Encoder Bootstrapping (ICLR 2025 · similarity 0.90)
- Grounded Object-Centric Learning (ICLR 2024 · similarity 0.90)
- GLASS: Guided Latent Slot Diffusion for Object-Centric Learning (CVPR 2025 · similarity 0.89)
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.