Research map / Image generation / Fast diffusion sampling
On Analyzing Generative and Denoising Capabilities of Diffusion-based Deep Generative Models
Kamil Deja, Anna Kuzina, Tomasz Trzcinski, Jakub Mikolaj Tomczak
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 first diffusion steps with a separate denoising autoencoder (DAED), VLB loss (ablation) | DDGM · CIFAR10 | FID: 26 -> 15 | Improved |
| replace first diffusion steps with a separate denoising autoencoder (DAED), VLB loss (ablation) | DDGM · CelebA | FID: 23 -> 16 | Improved |
| replace first diffusion steps with DAED, VLB loss (ablation) | DDGM · FashionMNIST | FID: 8.9 -> 7.5 | Improved |
| replace first diffusion steps with DAED, simplified loss (ablation) | DDGM · CIFAR10 | FID: 7.2 -> 14.8 | Hurt |
| replace first diffusion steps with DAED, simplified loss (ablation) | DDGM · CelebA | FID: 4.9 -> 7.4 | Hurt |
| replace first diffusion steps with DAED, simplified loss (ablation) | DDGM · FashionMNIST | FID: 7.8 -> 5.7 | Improved |
| increase noise handled by DAE (beta1 0.001 -> 0.1) (ablation) | DAED · CelebA, simplified loss | FID: 7.4 -> 22 | Hurt |
| use DAED instead of DDGM for denoising out-of-domain images | DDGM trained on CIFAR10 · transfer to CIFAR100 | MAE: 0.097 -> 0.074 | Improved |
| use DAED instead of DDGM for denoising out-of-domain images | DDGM trained on ImageNet · transfer to CelebA | MAE: 0.068 -> 0.050 | Improved |
| replace up to 10% of DDGM steps with DAE (ablation) | DDGM (simplified objective) · FashionMNIST, CIFAR10, CelebA | FID | No clear effect |
Closest papers by meaning
- Denoising Diffusion Bridge Models (ICLR 2024 · similarity 0.91)
- Compressed Image Generation with Denoising Diffusion Codebook Models (ICML 2025 · similarity 0.91)
- Think while You Generate: Discrete Diffusion with Planned Denoising (ICLR 2025 · similarity 0.90)
- Diffusion Bridge Implicit Models (ICLR 2025 · similarity 0.90)
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference (NeurIPS 2024 · similarity 0.90)
- Directly Denoising Diffusion Models (ICML 2024 · similarity 0.90)
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.
