Research map / Detection and segmentation / Open-set and few-shot detection
Unsupervised Layered Image Decomposition Into Object Prototypes
Tom Monnier, Elliot Vincent, Jean Ponce, Mathieu Aubry
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 learned occlusion prediction (vs fixed occlusion) (ablation) | DTI-Sprites · Multi-dSprites | ARI-FG: 85.7 -> 91.5 | Improved |
| add learned occlusion prediction (vs fixed occlusion) (ablation) | DTI-Sprites · Multi-dSprites | ARI: 94.2 -> 95.0 | Improved |
| add shared transformations besides sprite-specific ones (ablation) | DTI-Sprites · Tetrominoes | ARI-FG: 95.3 -> 99.6 | Improved |
| add shared transformations besides sprite-specific ones (ablation) | DTI-Sprites · Tetrominoes | ARI: 82.6 -> 99.8 | Improved |
| use greedy sprite selection instead of exhaustive search (ablation) | DTI-Sprites · Multi-dSprites2 (2-object subset) | ARI-FG: 94.4 -> 95.5 | No clear effect |
| use greedy sprite selection instead of exhaustive search (ablation) | DTI-Sprites · Multi-dSprites2 (2-object subset) | ARI: 95.9 -> 95.2 | No clear effect |
| increase lambda from 1e-5 to 1e-4 (ablation) | DTI-Sprites · Multi-dSprites | qualitative reconstruction | Improved |
| increase lambda to 1e-3 (ablation) | DTI-Sprites · Multi-dSprites | qualitative reconstruction | Hurt |
| use more sprites than ground-truth classes (ablation) | DTI-Sprites · Instagram collections | qualitative sprites | No clear effect |
| use pixel-space clustering with sprites instead of DTI-Clustering | DTI-Sprites · SVHN | clustering accuracy: 57.4 -> 63.1 | Improved |
| use sprites instead of DTI-Clustering | DTI-Sprites · GTSRB-8 | clustering accuracy: 54.3 -> 89.4 | Improved |
Closest papers by meaning
- Cycle-Consistent Tuning for Layered Image Decomposition (CVPR 2026 · similarity 0.89)
- CObL: Toward Zero-Shot Ordinal Layering without User Prompting (ICCV 2025 · similarity 0.88)
- Associating Objects and Their Effects in Video through Coordination Games (NeurIPS 2022 · similarity 0.87)
- "Double-DIP": Unsupervised Image Decomposition via Coupled Deep-Image-Priors (CVPR 2019 · similarity 0.87)
- Generative Omnimatte: Learning to Decompose Video into Layers (CVPR 2025 · similarity 0.87)
- Compositional Image Decomposition with Diffusion Models (ICML 2024 · 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.
