Research map / Image generation / Fast diffusion sampling
Learning Generative ConvNets via Multi-Grid Modeling and Sampling
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 |
|---|---|---|---|
| Use multi-grid modeling and coarse-to-fine Langevin sampling instead of single-grid learning | Generative ConvNet · SVHN test, 1,000 labeled images | Test error rate: 36.69 -> 30.23 | Helped |
| Use multi-grid modeling and coarse-to-fine Langevin sampling instead of single-grid learning | Generative ConvNet · SVHN test, 2,000 labeled images | Test error rate: 30.87 -> 26.54 | Helped |
| Use multi-grid modeling and coarse-to-fine Langevin sampling instead of single-grid learning | Generative ConvNet · SVHN test, 4,000 labeled images | Test error rate: 25.60 -> 22.83 | Helped |
| Use multi-grid learned features with a CNN classifier rather than training the same classifier from scratch | SVHN classifier · SVHN test, 1,000 labeled images | Test error rate: 39.04 -> 19.73 | Helped |
| Use multi-grid learned features with a CNN classifier rather than training the same classifier from scratch | SVHN classifier · SVHN test, 2,000 labeled images | Test error rate: 22.26 -> 15.86 | Helped |
| Use multi-grid learned features with a CNN classifier rather than training the same classifier from scratch | SVHN classifier · SVHN test, 4,000 labeled images | Test error rate: 15.24 -> 12.71 | Helped |
| Use multi-grid modeling and sampling instead of DCGAN | CIFAR-10 generative model · CIFAR-10 | Inception score: 6.581 -> 6.565 | No clear effect |
| Use multi-grid inpainting instead of ContextEncoder | CelebA inpainting · Square mask | PSNR: 17.37 -> 16.42 | Hurt |
| Use multi-grid inpainting instead of ContextEncoder | CelebA inpainting · Doodle mask | PSNR: 15.40 -> 16.98 | Helped |
| Use multi-grid inpainting instead of ContextEncoder | CelebA inpainting · Pepper mask | PSNR: 17.04 -> 19.34 | Helped |
| Initialize sampling from the 1×1 training-image version and propagate samples from coarse to fine grids instead of initializing CD from observed images | Generative ConvNet · Image synthesis | Synthesis quality | Helped |
Closest papers by meaning
- A Theory of Generative ConvNet (ICML 2016 · similarity 0.91)
- Patchwise Generative ConvNet: Training Energy-Based Models From a Single Natural Image for Internal Learning (CVPR 2021 · similarity 0.90)
- Learning Energy-Based Generative Models via Coarse-to-Fine Expanding and Sampling (ICLR 2021 · similarity 0.86)
- Learning FRAME Models Using CNN Filters (AAAI 2016 · similarity 0.86)
- DiCo: Revitalizing ConvNets for Scalable and Efficient Diffusion Modeling (NeurIPS 2025 · similarity 0.84)
- Learning Energy-Based Models by Cooperative Diffusion Recovery Likelihood (ICLR 2024 · similarity 0.84)
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.