Research map / Image generation / Autoregressive and latent generation
Modeling Image Composition for Complex Scene Generation
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 |
|---|---|---|---|
| Use focal attention instead of the strongest reported baseline | TwFA · COCO-stuff | FID: 29.56 -> 22.15 | Helped |
| Use focal attention instead of the strongest reported baseline | TwFA · Visual Genome | FID: 19.14 -> 17.74 | Helped |
| Replace focal attention with global self-attention (Grid16) (ablation) | TwFA · COCO-stuff | FID: 22.15 -> 27.04 | Hurt |
| Use sliding-window attention with window size 2 (ablation) | TwFA · COCO-stuff | FID: 22.15 -> 27.64 | Hurt |
| Use sliding-window attention with window size 4 (ablation) | TwFA · COCO-stuff | FID: 22.15 -> 29.01 | Hurt |
| Add object-object interaction (ablation) | TwFA · COCO-stuff | AP50: 25.01% -> 28.20% | Helped |
| Add object-patch interaction (ablation) | TwFA · COCO-stuff | AP50: 27.59% -> 28.20% | Helped |
| Add patch-patch interaction (ablation) | TwFA · COCO-stuff | AP50: 26.00% -> 28.20% | Helped |
| Add all focal-attention interactions instead of global attention (ablation) | TwFA · COCO-stuff | AP75: 17.45% -> 20.12% (+15.3%) | Helped |
| Use TwFA rather than Grid16* for few-shot adaptation | Few-shot L2I · COCO-stuff novel zebra, 20 shots | FID: 39.47 -> 36.34 | Helped |
| Use TwFA rather than Grid16* for few-shot adaptation | Few-shot L2I · COCO-stuff novel zebra, 30 shots | FID: 39.32 -> 34.87 | Helped |
| Increase few-shot training examples from 20 to 50 | TwFA* · COCO-stuff novel zebra | FID: 36.34 -> 31.53 | Helped |
Closest papers by meaning
- Generating compositional scenes via Text-to-image RGBA Instance Generation (NeurIPS 2024 · similarity 0.88)
- Scene Graph Disentanglement and Composition for Generalizable Complex Image Generation (NeurIPS 2024 · similarity 0.88)
- R&B: Region and Boundary Aware Zero-shot Grounded Text-to-image Generation (ICLR 2024 · similarity 0.86)
- Compositional Scene Understanding through Inverse Generative Modeling (ICML 2025 · similarity 0.86)
- Enhancing High-Resolution 3D Generation through Pixel-wise Gradient Clipping (ICLR 2024 · similarity 0.85)
- A 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale (NeurIPS 2026 · similarity 0.85)
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.