Research map / Image generation / Fast diffusion sampling
Revisiting Spectral Representations in Generative Diffusion Models
Yuehao Wang, Peihao Wang, Hanwen Jiang, Ziyi Yang, Qixing Huang, Zhangyang Wang
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 |
|---|---|---|---|
| add spectral representation regularizer to diffusion training | DiT-L/4 · ImageNet 64x64 | FID: 9.441 -> 7.994 | Improved |
| add spectral regularizer | DiT-XL/2 latent · ImageNet 256x256 | FID: 2.508 -> 2.298 | Improved |
| add spectral regularizer | DiT-S/2 · CIFAR10 | FID: 11.588 -> 8.742 | Improved |
| add spectral regularizer | DiT-S/2 · CelebA | FID: 28.806 -> 25.678 | Improved |
| add spectral regularizer | DiT-S/2 · FFHQ 64x64 | FID: 13.766 -> 13.074 | Improved |
| add spectral regularizer | DiT-L/4 · ImageNet 64x64 | Inception Score: 102.069 -> 78.366 | Hurt |
| use external pretrained-encoder alignment (REPA) instead of self-supervised spectral alignment | DiT-XL/2 · ImageNet 256x256 latent | FID: 2.298 -> 1.745 | Improved |
| add spectral regularizer | DiT-3D-S/4 · ShapeNet Chair, 5K iters | 1-NNA (CD): 0.850 -> 0.583 | Improved |
| add spectral regularizer | DiT-3D-S/4 · ShapeNet Airplane, 10K iters | 1-NNA (CD): 0.852 -> 0.607 | Improved |
| add spectral regularizer | DiT-3D-S/4 · ShapeNet Chair, 5K iters | COV (CD): 0.295 -> 0.488 | Improved |
| add spectral regularizer | DiT-S/2 · training cost | training time: +10% (ImageNet-64), +70% (ImageNet-256), +20% (3D) | Hurt |
Closest papers by meaning
- Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Think (ICLR 2025 · similarity 0.91)
- Learning Diffusion Models with Flexible Representation Guidance (NeurIPS 2025 · similarity 0.90)
- An Analytical Theory of Spectral Bias in the Learning Dynamics of Diffusion Models (NeurIPS 2025 · similarity 0.90)
- Spectral Progressive Diffusion for Efficient Image and Video Generation (NeurIPS 2026 · similarity 0.88)
- Sparsely Supervised Diffusion (NeurIPS 2026 · similarity 0.88)
- CORAL: Disentangling Latent Representations in Long-Tailed Diffusion (NeurIPS 2025 · similarity 0.88)
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.
