Research map / Detection and segmentation / CNN features
Teaching Compositionality to CNNs
Check 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 the compositionality objective to standard CNN training | CNN object recognition · MS-COCO-sub | mAP: 25.5% | Helped |
| add the compositionality objective to standard CNN training | CNN object recognition · 3D-Single-Inst | top-k classification accuracy: 35.2% | Helped |
| add the compositionality objective to standard CNN training | CNN object recognition · MNIST-Single | top-k classification accuracy: 2.0% | Helped |
| add the compositionality objective to standard CNN training | CNN object recognition · MNIST-Multi | top-k classification accuracy: 2.6% | Helped |
| add dropout and l2-regularization to baseline-aug (ablation) | baseline-aug · MS-COCO-sub | mAP: 4.6% | Helped |
| add dropout and l2-regularization to baseline (ablation) | baseline · MS-COCO-sub | mAP: 4.1% | Helped |
| use object-only masks for the compositional penalty instead of excluding other objects (ablation) | comp-full · 3D-Single-Inst | top-k classification accuracy: 14.7% | Hurt |
| use object-only masks for the compositional penalty instead of excluding other objects (ablation) | comp-full · 3D-Single | top-k classification accuracy: 7.3% | Hurt |
| use object-only masks for the compositional penalty instead of excluding other objects (ablation) | comp-full · 3D-Multi-Inst | top-k classification accuracy: 4.4% | Hurt |
| use object-only masks for the compositional penalty instead of excluding other objects (ablation) | comp-full · 3D-Multi | top-k classification accuracy: 2.9% | Hurt |
| remove mask reapplication to masked CNN activations (ablation) | comp-full · 3D-Single-Inst | top-k classification accuracy: 0.3% | Hurt |
| remove mask reapplication to masked CNN activations (ablation) | comp-full · MNIST-Single | top-k classification accuracy: 20.4% | Hurt |
Closest papers by meaning
- Learning to Compose: Improving Object Centric Learning by Injecting Compositionality (ICLR 2024 · similarity 0.89)
- Scalable Evaluation and Neural Models for Compositional Generalization (NeurIPS 2025 · similarity 0.87)
- Does Data Scaling Lead to Visual Compositional Generalization? (ICML 2025 · similarity 0.87)
- Not Only Text: Exploring Compositionality of Visual Representations in Vision-Language Models (CVPR 2025 · similarity 0.86)
- Provable Compositional Generalization for Object-Centric Learning (ICLR 2024 · similarity 0.86)
- When does compositional structure yield compositional generalization? A kernel theory. (ICLR 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.