Research map / Image generation / Fast diffusion sampling
Beyond Generative Priors: Minority Sampling with JEPA-Guided Diffusion
Sol Park, Soobin Um
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 deferred guidance (tau 0.8 instead of 1.0) (ablation) | JEPA guidance · MS-COCO prompts, SDv1.5 | ImageReward: 0.11 -> 0.23 | Improved |
| add deferred guidance (tau 0.8 instead of 1.0) (ablation) | JEPA guidance · MS-COCO prompts, SDv1.5 | CLIPScore: 31.26 -> 31.40 | Improved |
| increase guidance step size eta (ablation) | JEPA guidance · MS-COCO prompts, SDv1.5 | JEPA-SCORE: -289.79 -> -344.76 | Improved |
| increase guidance step size eta (ablation) | JEPA guidance · MS-COCO prompts, SDv1.5 | CLIPScore: 31.44 -> 31.36 | Hurt |
| increase RSVD rank k from 9 to 15 (ablation) | JEPA guidance · MS-COCO prompts, SDXL-Lightning | JEPA-SCORE: -344.85 -> -335.28 | No clear effect |
| use JEPA guidance with different encoders (ViT-S, ViT-L, MetaCLIP) (ablation) | JEPA guidance · MS-COCO prompts, SDv1.5 | JEPA-SCORE: -289.79 -> -361.52 (ViT-S, eta 0.08) | Improved |
| replace generator-centric minority samplers with JEPA guidance | JEPA guidance · ImageNet 256x256 | cFID: 32.01 (BnS) -> 18.33 | Improved |
| replace generator-centric minority samplers with JEPA guidance | JEPA guidance · CelebA 64x64 | cFID: 58.27 (MG) -> 8.50 | Improved |
| add JEPA-guided minority samples (30K) to classifier training | ResNet-18 classifier · CelebA 40-attribute classification | F1: 0.746 -> 0.775 | Improved |
| add JEPA-guided minority samples (30K) to classifier training | ResNet-18 classifier · CelebA 40-attribute classification | accuracy: 0.898 -> 0.902 | No clear effect |
| JEPA guidance vs MinorityPrompt | JEPA guidance · MS-COCO prompts, SDv1.5 | ImageReward: 0.24 -> 0.22 | Hurt |
Closest papers by meaning
- Don't Play Favorites: Minority Guidance for Diffusion Models (ICLR 2024 · similarity 0.91)
- Boost-and-Skip: A Simple Guidance-Free Diffusion for Minority Generation (ICML 2025 · similarity 0.90)
- SPARKE: Scalable Prompt-Aware Diversity and Novelty Guidance in Diffusion Models via RKE Score (NeurIPS 2025 · similarity 0.88)
- Minority-Focused Text-to-Image Generation via Prompt Optimization (CVPR 2025 · similarity 0.88)
- Denoising with a Joint-Embedding Predictive Architecture (ICLR 2025 · similarity 0.87)
- Domain Guidance: A Simple Transfer Approach for a Pre-trained Diffusion Model (ICLR 2025 · 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.
