Research map / Image generation / Editing and style transfer
Efficient Spatially Sparse Inference for Conditional GANs and Diffusion Models
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 |
|---|---|---|---|
| Apply spatially sparse inference to DDPM | DDPM · LSUN Church | MACs: 249G -> 65.3G | Helped |
| Apply spatially sparse inference to DDPM | DDPM · LSUN Church | PSNR with G.T.: 26.8 -> 26.8 | No clear effect |
| Apply spatially sparse inference to GauGAN | GauGAN · Cityscapes | MACs: 281G -> 30.7G | Helped |
| Apply SSI to GAN Compression | GauGAN · Cityscapes | MACs: 9.64G -> 7.06G | Helped |
| Apply SSI with attention query pruning | Stable Diffusion · LAION-5B | MACs: 805G -> 387G | Helped |
| Use tiling-based sparse convolution (ablation) | DDPM · RTX 2080Ti | latency: 54.6ms -> 34.0ms | Helped |
| Add precomputed normalization parameters (ablation) | DDPM with sparse convolution · RTX 2080Ti | latency: 34.0ms -> 29.6ms | Helped |
| Add fusion of element-wise operations (ablation) | DDPM with sparse convolution and precomputed normalization · RTX 2080Ti | latency: 29.6ms -> 20.7ms | Helped |
| Add Scatter fusion (ablation) | DDPM with earlier SSI optimizations · RTX 2080Ti | latency: 20.7ms -> 19.1ms | Helped |
| Add SIGE attention layers (ablation) | Stable Diffusion with sparse convolution · RTX 3090 | latency: 335ms -> 76.4ms | Helped |
| Increase mask dilation (ablation) | GauGAN · Cityscapes | image quality at edited-region boundaries | Helped |
| Increase mask dilation (ablation) | GauGAN · Cityscapes | computation | Hurt |
Closest papers by meaning
- Accelerating Text-to-Image Editing via Cache-Enabled Sparse Diffusion Inference (AAAI 2024 · similarity 0.89)
- Exploring Sparse MoE in GANs for Text-conditioned Image Synthesis (CVPR 2025 · similarity 0.86)
- E$^2$GAN: Efficient Training of Efficient GANs for Image-to-Image Translation (ICML 2024 · similarity 0.86)
- SparseDiT: Token Sparsification for Efficient Diffusion Transformer (NeurIPS 2025 · similarity 0.86)
- Improved Training Technique for Shortcut Models (NeurIPS 2025 · similarity 0.86)
- Stochastic Interpolants via Conditional Dependent Coupling (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.