Research map / Image generation / Fast diffusion sampling
Dynamic Dual-Output Diffusion Models
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 |
|---|---|---|---|
| Replace the single prediction path with dynamically interpolated noise and image prediction | DDPM on CIFAR10 with a linear schedule · CIFAR10 32×32, 5 iterations | FID: 49.70 -> 35.12 | Helped |
| Replace the single prediction path with dynamically interpolated noise and image prediction | IDDPM with a cosine schedule · CIFAR10 32×32, 100 iterations | FID: 4.58 -> 5.10 | Hurt |
| Replace the single prediction path with dynamically interpolated noise and image prediction | DDPM on CelebA · CelebA 64×64, linear schedule, 5 iterations | FID: 56.16 -> 26.22 | Helped |
| Use dynamic r_θ instead of a fixed mean r_t (ablation) | Dual-output DDPM · CIFAR10 32×32, linear schedule, 5 iterations | FID: 38.50 -> 35.12 | Helped |
| Add the x_θ prediction path to a noise-only model (ablation) | Dual-output DDPM · CIFAR10 32×32, linear schedule, 5 iterations | FID: 41.99 -> 35.12 | Helped |
| Add the ε_θ prediction path to an image-only model (ablation) | Dual-output DDPM · CIFAR10 32×32, linear schedule, 5 iterations | FID: 45.53 -> 35.12 | Helped |
| Use dual-output prediction rather than either single path (ablation) | ImageNet 128×128 model · ImageNet, no guidance, 25 iterations | FID: 51.3 -> 27.7; 29.5 -> 27.7 | Helped |
| Use dual-output prediction rather than either single path (ablation) | ImageNet 128×128 model · ImageNet, classifier scale 1.0, 25 iterations | FID: 44.1 -> 24.5; 26.0 -> 24.5 | Helped |
| Prefer the x_θ path to ε_θ alone (ablation) | Dual-output model's single-path variants · CIFAR10 cosine schedule and CelebA linear schedule | FID | Mixed |
| Add classifier guidance at scale 1.0 | ImageNet dual-output model · ImageNet 128×128, 25 iterations | FID: 27.7 -> 24.5 | Helped |
Closest papers by meaning
- Denoising Diffusion Bridge Models (ICLR 2024 · similarity 0.91)
- You Only Sample Once: Taming One-Step Text-to-Image Synthesis by Self-Cooperative Diffusion GANs (ICLR 2025 · similarity 0.90)
- Evaluating the design space of diffusion-based generative models (NeurIPS 2024 · similarity 0.90)
- Directly Denoising Diffusion Models (ICML 2024 · similarity 0.90)
- NoiseCtrl: A Sampling-Algorithm-Agnostic Conditional Generation Method for Diffusion Models (CVPR 2025 · similarity 0.90)
- Denoising Autoregressive Transformers for Scalable Text-to-Image Generation (ICLR 2025 · 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.