Research map / Detection and segmentation / Open-set and few-shot detection
Object-Centric Learning with Slot Attention
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 |
|---|---|---|---|
| Use iterative slot attention instead of the soft k-means variant (ablation) | Slot Attention · CLEVR6 | ARI: 75.5\pm3.8\% -> 98.8\pm0.3\% | Helped |
| Use weighted mean rather than weighted sum aggregation (ablation) | Slot Attention · CLEVR6 and CLEVR10 | ARI and AP | Helped |
| Add position embeddings to CNN feature maps (ablation) | Slot Attention · CLEVR6 object discovery and CLEVR10 position prediction | ARI and position prediction | Helped |
| Use shared rather than per-slot Gaussian initialization parameters (ablation) | Slot Attention · CLEVR6 object discovery and supervised property prediction | ARI and property prediction | Mixed |
| Normalize attention over slots rather than inputs (ablation) | Slot Attention · CLEVR6 and CLEVR10 | ARI and property prediction | Helped |
| Use the learned GRU update rather than directly using aggregated updates as slots (ablation) | Slot Attention · CLEVR6 and CLEVR10 | ARI and property prediction | Helped |
| Use more than one attention iteration during training (ablation) | Slot Attention · CLEVR6 and CLEVR10 | ARI and property prediction | Helped |
| Train with more than 3 attention iterations for the same number of steps (ablation) | Slot Attention · CLEVR6 and CLEVR10 | ARI and property prediction | Hurt |
| Apply LayerNorm to inputs and slot representations (ablation) | Slot Attention · CLEVR6 and CLEVR10 | ARI and property prediction | Helped |
| Remove the residual MLP after the GRU (ablation) | Slot Attention · CLEVR6 object discovery and CLEVR10 property prediction | ARI and property prediction | Mixed |
| Scale the softmax temperature by sqrt(D) (ablation) | Slot Attention · CLEVR6 and CLEVR10 | ARI and property prediction | Helped |
| Use the attention-map offset rather than putting the offset in the weighted-mean denominator (ablation) | Slot Attention · CLEVR6 and CLEVR10 | ARI and property prediction | No clear effect |
Closest papers by meaning
- Grounded Object-Centric Learning (ICLR 2024 · similarity 0.93)
- Adaptive Slot Attention: Object Discovery with Dynamic Slot Number (CVPR 2024 · similarity 0.93)
- GLASS: Guided Latent Slot Diffusion for Object-Centric Learning (CVPR 2025 · similarity 0.92)
- MetaSlot: Break Through the Fixed Number of Slots in Object-Centric Learning (NeurIPS 2025 · similarity 0.91)
- Identifiable Object-Centric Representation Learning via Probabilistic Slot Attention (NeurIPS 2024 · similarity 0.91)
- Slot-Guided Adaptation of Pre-trained Diffusion Models for Object-Centric Learning and Compositional Generation (ICLR 2025 · similarity 0.91)
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.