Research map / Image generation / Fast diffusion sampling
Grid Distillation: Compositional Image Distillation via Structured Generative Grids
Biplab Ch Das, Shouvik Das, Viswanath Gopalakrishnan
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 |
|---|---|---|---|
| replace bilinear upsampling with single-step diffusion detail enhancement (ablation) | Grid-Distil · ImageNet-1K (IPC 10) | top-1 accuracy: 35.40 -> 50.01 | Improved |
| replace bilinear upsampling with diffusion detail enhancement (ablation) | Grid-Distil · ImageNette (ResNetAP-10) | top-1 accuracy: 81.5 -> 83.3 (IPC 10); 91.5 -> 92.7 (IPC 50) | Improved |
| replace bilinear upsampling with diffusion detail enhancement (ablation) | Grid-Distil · ImageWoof (ResNetAP-10) | top-1 accuracy: 62.5 -> 63.9 (IPC 10); 82.2 -> 82.4 (IPC 50) | Improved |
| replace bilinear upsampling with diffusion detail enhancement (ablation) | Grid-Distil · ImageIDC (ResNetAP-10) | top-1 accuracy: 74.7 -> 73.5 (IPC 10); 87.7 -> 87.1 (IPC 50) | No clear effect |
| add coverage term (alpha=1.0) alone to grid selection (ablation) | Grid selection objective · Real ImageWoof grids (IPC 20) | top-1 accuracy: 74.6 -> 78.2 | Improved |
| add diversity term (beta=1.0) alone to grid selection (ablation) | Grid selection objective · Real ImageWoof grids (IPC 20) | top-1 accuracy: 74.6 -> 74.4 | No clear effect |
| add spectral term (gamma=1.0) alone to grid selection (ablation) | Grid selection objective · Real ImageWoof grids (IPC 20) | top-1 accuracy: 74.6 -> 75.8 | Improved |
| combine all three terms (alpha=1.0, beta=0.6, gamma=0.3) vs coverage alone (ablation) | Grid selection objective · Real ImageWoof grids (IPC 20) | top-1 accuracy: 78.2 -> 78.6 | Improved |
| add action labels to grid enhancement prompts (ablation) | Grid-LDC · UCF-101 Mid-Frames (few-shot) | few-shot accuracy: 71.66 -> 73.88 (1-shot); 78.01 -> 79.03 (8-shot) | Improved |
| apply Grid Distillation on top of LDC | LDC · UCF-101 Mid-Frames (few-shot) | few-shot accuracy: 66.83 -> 71.66 (1-shot); 77.03 -> 78.01 (8-shot) | Improved |
| use Grid-Distil instead of VLCP | dataset distillation on ImageWoof · ImageWoof (IPC 50, ResNet-18) | top-1 accuracy: 58.9 -> 84.3 | Improved |
Closest papers by meaning
- Efficient Dataset Distillation via Minimax Diffusion (CVPR 2024 · similarity 0.87)
- Unlocking Dataset Distillation with Diffusion Models (NeurIPS 2025 · similarity 0.87)
- MGD$^3$ : Mode-Guided Dataset Distillation using Diffusion Models (ICML 2025 · similarity 0.86)
- Improved Distribution Matching Distillation for Fast Image Synthesis (NeurIPS 2024 · similarity 0.86)
- D^4: Dataset Distillation via Disentangled Diffusion Model (CVPR 2024 · similarity 0.86)
- RaPD: Resolution-Agnostic Pixel Diffusion via Semantics-Enriched Implicit Representations (NeurIPS 2026 · 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.
