Research map / Image generation / Fast diffusion sampling
Improving Diffusion Models for Class-imbalanced Training Data via Capacity Manipulation
Feng Hong, Jiangchao Yao, Yifei Shen, Dongsheng Li, Ya Zhang, Yanfeng 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 Capacity Manipulation | DDPM · Imb. CIFAR-100 (IR=100) | FID: 10.163 -> 7.519 | Improved |
| add Capacity Manipulation | DDPM · Imb. CIFAR-10 (IR=100) | FID: 10.697 -> 7.727 | Improved |
| add Capacity Manipulation | DDPM · Imb. CelebA-HQ (IR=100) male | FID: 16.425 -> 12.788 | Improved |
| add Capacity Manipulation | DDPM · Imb. iNaturalist 64x64 | FID: 7.99 -> 6.82 | Improved |
| add Capacity Manipulation | Stable Diffusion LoRA · Imb. ArtBench-10 (IR=100) | FID: 27.083 -> 22.776 | Improved |
| add consistency loss L_Con (ablation) | CM · Imb. CIFAR-100 (IR=100) | FID: 8.412 -> 7.519 | Improved |
| add diversity loss L_Div (ablation) | CM · Imb. CIFAR-100 (IR=100) | FID: 8.073 -> 7.519 | Improved |
| add reserved minority capacity theta_e at inference (vs theta_g only) (ablation) | CM · Imb. CIFAR-100 (IR=100) Few split | Few FID: 29.357 -> 18.729 | Improved |
| add reserved minority capacity theta_e at inference (vs theta_g only) (ablation) | CM · Imb. CIFAR-100 (IR=100) overall | FID: 13.712 -> 7.519 | Improved |
| use low-rank decomposition with CBDM objective only | CBDM · Imb. CIFAR-100 (IR=100) | FID: 10.051 -> 10.231 | No clear effect |
| fine-tune CBDM with LoRA instead of CM | CBDM · Imb. CIFAR-100 (IR=100) | FID: 10.051 -> 11.424 | Hurt |
| use CM instead of Group-Expert LoRA (MoE-style) (ablation) | LoRA adapters · Imb. CIFAR-100 | FID: 10.06 -> 7.519 | Improved |
Closest papers by meaning
- CORAL: Disentangling Latent Representations in Long-Tailed Diffusion (NeurIPS 2025 · similarity 0.87)
- Analyzing and Improving the Training Dynamics of Diffusion Models (CVPR 2024 · similarity 0.87)
- Slight Corruption in Pre-training Data Makes Better Diffusion Models (NeurIPS 2024 · similarity 0.87)
- Multi-Architecture Multi-Expert Diffusion Models (AAAI 2024 · similarity 0.87)
- Don't Play Favorites: Minority Guidance for Diffusion Models (ICLR 2024 · similarity 0.87)
- On improved Conditioning Mechanisms and Pre-training Strategies for Diffusion Models (NeurIPS 2024 · similarity 0.87)
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.
