Research map / Detection and segmentation / Open-set and few-shot detection
COAT: Measuring Object Compositionality in Emergent Representations
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 |
|---|---|---|---|
| remove duplicate slots (replace with mean of other slots) in slot attention (ablation) | Slot Attention · CLEVR (w/ occlusion) | L2-COAT: 48.55 -> 60.70 | Helped |
| remove invisible slots using decoder mask weights, on top of duplicate removal (ablation) | Slot Attention · CLEVR (w/ occlusion) | L2-COAT: 60.70 -> 77.07 | Helped |
| train on correlated, cluttered same-color/material object dataset (ablation) | Slot Attention · CLEVR (correlated train set) | L2-COAT: 77.07 -> 64.82 | Hurt |
| use a factorized prior (IODINE) with slot structure instead of generic slot attention | IODINE · CLEVR (w/ occlusion) | L2-COAT: 47.52 vs 48.55 for slot attention | No clear effect |
| train on correlated object dataset (ablation) | IODINE · CLEVR (correlated train set) | acos-COAT: 16.33 -> 9.16 (and fails pixel-level hard negative test in L2) | Hurt |
| use an autoencoder or beta-TC-VAE (no slot structure) instead of slot-based models | Autoencoder / beta-TC-VAE · CLEVR (w/ occlusion) | HN test / COAT: Fail (no COAT score) | Hurt |
| use MoCo v2 self-supervised ConvNet instead of generative slot models | MoCo v2 ConvNet · CLEVR (w/ occlusion) | acos-COAT: 14.05 vs pixel baseline 36.28 | Hurt |
| evaluate raw slot attention representations without post-processing | Slot Attention · CLEVR (w/ occlusion) | L2-COAT: 48.55 vs pixel baseline 75.47 | Hurt |
Closest papers by meaning
- Learning to Compose: Improving Object Centric Learning by Injecting Compositionality (ICLR 2024 · similarity 0.89)
- Compositional Scene Understanding through Inverse Generative Modeling (ICML 2025 · similarity 0.87)
- Provable Compositional Generalization for Object-Centric Learning (ICLR 2024 · similarity 0.87)
- Unlocking Compositional Generalization in Continual Few-Shot Learning (NeurIPS 2026 · similarity 0.87)
- Not Only Text: Exploring Compositionality of Visual Representations in Vision-Language Models (CVPR 2025 · similarity 0.86)
- Does Data Scaling Lead to Visual Compositional Generalization? (ICML 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.