Research map / Image generation / Fast diffusion sampling
Adversarial score matching and improved sampling for image generation
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 |
|---|---|---|---|
| Denoise the final Langevin sample with the score network (ablation) | DSM-ALS · CIFAR-10 and LSUN-Churches | FID: 36.3 / 13.3; 43.2 / 40.3 | Helped |
| Use consistent instead of non-consistent sampling (ablation) | DSM, main score network architecture · CIFAR-10, nσ=1 | FID (non-denoised): 36.3 -> 14.7 | Helped |
| Use consistent instead of non-consistent sampling (ablation) | DSM, unconditional score network architecture · CIFAR-10, nσ=5, adversarial, denoised | FID: 6.1 -> 6.5 | Mixed |
| Add adversarial training (ablation) | DSM, main score network architecture · CIFAR-10, non-consistent sampling, nσ=5 | FID (denoised): 10.9 -> 9.5 | Helped |
| Add adversarial training (ablation) | DSM, unconditional score network architecture · CIFAR-10, non-consistent sampling, nσ=5 | FID (denoised): 5.6 -> 6.1 | Hurt |
| Increase nσ from 1 to 5 (ablation) | DSM, main score network architecture · CIFAR-10, non-adversarial non-consistent sampling | FID: 36.3 -> 33.7 | Helped |
| Increase nσ from 1 to 5 (ablation) | DSM, main score network architecture · LSUN-Churches, non-adversarial non-consistent sampling | FID (denoised): 40.3 -> 39.2 | Helped |
| Add adversarial training (ablation) | DSM, main score network architecture · LSUN-Churches, consistent sampling, nσ=5 | FID (denoised): 39.1 -> 35.4 | Helped |
| Add adversarial training (ablation) | DSM, unconditional score network architecture · 3-Stacked MNIST | Modes (maximum 1000): 1000.0 -> 1000.0 | No clear effect |
| Add adversarial training (ablation) | DSM, diffusion-network architecture · CIFAR-10, 50k samples | FID (denoised): 3.65 -> 4.02 | Hurt |
Closest papers by meaning
- SMaRt: Improving GANs with Score Matching Regularity (ICML 2024 · similarity 0.88)
- Adversarial Score Distillation: When score distillation meets GAN (CVPR 2024 · similarity 0.86)
- Evaluating the design space of diffusion-based generative models (NeurIPS 2024 · similarity 0.86)
- Neural Network-Based Score Estimation in Diffusion Models: Optimization and Generalization (ICLR 2024 · similarity 0.86)
- Improving the Training of Data-Efficient GANs via Quality Aware Dynamic Discriminator Rejection Sampling (CVPR 2025 · similarity 0.86)
- Score Distillation via Reparametrized DDIM (NeurIPS 2024 · 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.