Research map / Detection and segmentation / Open-set and few-shot detection
Unlocking Slot Attention by Changing Optimal Transport Costs
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 MESH entropy minimization in place of standard slot attention | Slot attention · Random objects detection, σ=0.01 | normalized RMSE: 0.65 -> 0.31 | Helped |
| Use MESH entropy minimization rather than Sinkhorn alone | Slot attention · Multi-dSprites | FG-ARI: 87.2 ± 1.8 -> 95.6 ± 0.2 | Helped |
| Use MESH entropy minimization rather than Sinkhorn alone | Slot attention · ClevrTex | FG-mIoU: 35.3 ± 4.4 -> 43.2 ± 4.0 | Helped |
| Use MESH entropy minimization rather than Sinkhorn alone | Slot attention · CLEVRER-L | FG-mIoU: 7.0 ± 0.2 -> 54.4 ± 8.9 | Helped |
| Use MESH entropy minimization rather than Sinkhorn alone | Slot attention · CLEVRER-S | TC: 89.7 ± 2.4 -> 80.2 ± 10.1 | Hurt |
| Replace regularized transport with EMD | Slot attention · Random objects detection | normalized RMSE: 0.23 vs 1.04 | Mixed |
| Increase slot attention iterations from 3 to 5 (ablation) | Slot attention on Multi-dSprites · Multi-dSprites | FG-ARI: 92.2±0.5 -> 81.9±6.1 | Hurt |
| Learn the slot initialization (ablation) | Slot attention on Multi-dSprites · Multi-dSprites | FG-ARI: 92.2±0.5 -> 93.0±1.0 | No clear effect |
| Reduce the MESH learning rate (ablation) | SA-MESH · CLEVRER-S | TC: 95.1% | Helped |
| Use learned slot noise instead of MESH (ablation) | Slot attention on CLEVRER-L · CLEVRER-L | FG-mIoU: 54.4 ± 8.9 vs 10.2 ± 1.2 | Hurt |
Closest papers by meaning
- Adaptive Slot Attention: Object Discovery with Dynamic Slot Number (CVPR 2024 · similarity 0.91)
- GLASS: Guided Latent Slot Diffusion for Object-Centric Learning (CVPR 2025 · similarity 0.90)
- Grounded Object-Centric Learning (ICLR 2024 · similarity 0.89)
- MetaSlot: Break Through the Fixed Number of Slots in Object-Centric Learning (NeurIPS 2025 · similarity 0.89)
- Slot-Guided Adaptation of Pre-trained Diffusion Models for Object-Centric Learning and Compositional Generation (ICLR 2025 · similarity 0.88)
- Identifiable Object-Centric Representation Learning via Probabilistic Slot Attention (NeurIPS 2024 · similarity 0.88)
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.