Research map / Detection and segmentation / Open-set and few-shot detection
Improving Object-centric Learning with Query Optimization
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 |
|---|---|---|---|
| replace sampled slot initialization with learnable queries and use bi-level optimization with straight-through gradients | Slot-Attention · ShapeStacks | ARI-FG: 86.7 -> 92.9 | Helped |
| replace sampled slot initialization with learnable queries and use bi-level optimization with straight-through gradients | Slot-Attention · Stanford Dogs | IoU: 71.0 -> 82.5 | Helped |
| use a transformer decoder instead of a mixture decoder | BO-QSA · ShapeStacks | ARI-FG: 0.93$\pm$0.01 -> 0.68$\pm$0.02 | Hurt |
| use a transformer decoder instead of a mixture decoder | BO-QSA · Stanford Dogs | IoU: 36.8 -> 82.5 | Helped |
| use learnable queries rather than sampled initialization (ablation) | Slot-Attention · Stanford Dogs | IoU: 71.0 -> 64.5 | Hurt |
| use learnable queries rather than sampled initialization (ablation) | Slot-Attention · ShapeStacks | ARI-FG (%): 86.7 -> 88.1 | Helped |
| apply I-SA gradient detachment to query-initialized Slot-Attention without straight-through gradients (ablation) | QSA · Stanford Dogs | IoU: 64.5 -> 59.3 | Hurt |
| apply I-SA gradient detachment to query-initialized Slot-Attention without straight-through gradients (ablation) | QSA · ShapeStacks | ARI-FG (%): 88.1 -> 84.6 | Hurt |
| add straight-through gradients to initialization queries while detaching iterative refinement gradients (ablation) | I-QSA · Stanford Dogs | IoU: 59.3 -> 82.5 | Helped |
| add straight-through gradients to initialization queries while detaching iterative refinement gradients (ablation) | I-QSA · ShapeStacks | ARI-FG (%): 84.6 -> 92.9 | Helped |
| increase I-QSA training iterations from 3 to 7 (ablation) | I-QSA · Stanford Dogs | IoU: 59.3 -> 80.5 | Helped |
| update initialization queries with running means of post-iteration slots (ablation) | BO-QSA · ShapeStacks | ARI-FG: 92.9 -> 7.5 | Hurt |
Closest papers by meaning
- Adaptive Slot Attention: Object Discovery with Dynamic Slot Number (CVPR 2024 · similarity 0.92)
- Grounded Object-Centric Learning (ICLR 2024 · similarity 0.91)
- Identifiable Object-Centric Representation Learning via Probabilistic Slot Attention (NeurIPS 2024 · similarity 0.91)
- GLASS: Guided Latent Slot Diffusion for Object-Centric Learning (CVPR 2025 · similarity 0.90)
- MetaSlot: Break Through the Fixed Number of Slots in Object-Centric Learning (NeurIPS 2025 · similarity 0.90)
- SPOT: Self-Training with Patch-Order Permutation for Object-Centric Learning with Autoregressive Transformers (CVPR 2024 · similarity 0.90)
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.