Research map / Image generation / Fast diffusion sampling
Contrastive Conditional–Unconditional Alignment for Long-tailed Diffusion Model
Fang Chen, Alex Villa, Gongbo Liang, Li Fuxin, Xiaoyi Lu, Meng Tang
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 CCUA losses | SiT · ImageNet-LT 900k | FID: 19.9 -> 15.1 | Improved |
| add CCUA losses | SiT · ImageNet-LT 900k | FID_tail: 33.9 -> 22.5 | Improved |
| add CCUA losses | SiT · ImageNet-LT 900k | Recall (%): 18.6 -> 17.3 | Hurt |
| add CCUA losses | DDPM · TinyImageNet-LT | FID_tail: 40.1 -> 30.4 | Improved |
| add CCUA losses | DDPM · Places-LT | FID: 13.9 -> 12.0 | Improved |
| add CCUA losses | DDPM · CIFAR10-LT (DDIM 100 steps) | FID: 5.93 -> 5.56 | Improved |
| add CBDM | DDPM · TinyImageNet-LT | FID: 18.7 -> 20.9 | Hurt |
| add Dispersive Loss (direct conditional diversification) | SiT · ImageNet-LT 250k | FID: 33.8 -> 34.0 | No clear effect |
| add batch resampling (ablation) | DDPM · TinyImageNet-LT | FID_tail: 40.1 -> 33.6 | Improved |
| add batch resampling (ablation) | CCUA · TinyImageNet-LT | FID: 17.2 -> 15.2 | Improved |
| add batch resampling (ablation) | SiT · ImageNet-LT 900k | Recall (%): 18.6 -> 16.6 | Hurt |
| add alignment loss to unconditional contrastive loss (alpha 0.05, gamma 0.05 vs alpha 0.05 only) (ablation) | CCUA · ImageNet-LT 250k | FID: 27.7 -> 27.0 | Improved |
Closest papers by meaning
- CORAL: Disentangling Latent Representations in Long-Tailed Diffusion (NeurIPS 2025 · similarity 0.88)
- A Simple Approach to Unifying Diffusion-based Conditional Generation (ICLR 2025 · similarity 0.87)
- DEFT: Efficient Fine-tuning of Diffusion Models by Learning the Generalised $h$-transform (NeurIPS 2024 · similarity 0.86)
- Referee Can Play: An Alternative Approach to Conditional Generation via Model Inversion (ICML 2024 · similarity 0.86)
- Your contrastive learning problem is secretly a distribution alignment problem (NeurIPS 2024 · similarity 0.86)
- Stochastic Conditional Diffusion Models for Robust Semantic Image Synthesis (ICML 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.
