Research map / Image generation / Fast diffusion sampling
EGC: Image Generation and Classification via a Diffusion Energy-Based Model
Qiushan Guo, Chuofan Ma, Yi Jiang, Zehuan Yuan, Yizhou Yu, Ping Luo
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 classifier (joint distribution) to unsupervised EGC (ablation) | unsupervised EGC · CIFAR-10 | FID: 5.36 -> 3.49 | Improved |
| add classifier guidance in sampling (ablation) | EGC · CIFAR-10 | FID: 3.49 -> 3.30 | Improved |
| use U-Net instead of standard feedforward ResNet (ablation) | EGC · CIFAR-10 | FID: 7.15 -> 3.30 | Improved |
| add RandResizeCrop augmentation | EGC · ImageNet-1k 256x256 | accuracy: 72.5 -> 78.9 | Improved |
| jointly train conditional and unconditional model | EGC · ImageNet-1k 256x256 | FID: 17.5 -> 6.77 | Improved |
| use EGC over JEM for hybrid modeling | hybrid classifier-generator · CIFAR-10 | FID: 38.4 -> 3.30 | Improved |
| EGC single network vs Wide ResNet-28-12 classifier on CIFAR-10 | EGC · CIFAR-10 | accuracy: 95.6 -> 95.9 | Improved |
| EGC single network vs Wide ResNet-28-12 classifier on CIFAR-100 | EGC · CIFAR-100 | accuracy: 79.5 -> 77.9 | Hurt |
| EGC vs DDPM generation | EGC · CIFAR-10 and LSUN Church | FID: 3.17 -> 3.30 (CIFAR-10); 7.89 -> 8.97 (LSUN Church) | Hurt |
| increase guidance scale from 2 to 200 (ablation) | EGC conditional sampling · conditional sampling | diversity | Hurt |
| train EGC classifier through diffusion energy-based model | EGC · CIFAR-10 adversarial attacks | adversarial robustness | Improved |
Closest papers by meaning
- Learning Energy-Based Models by Cooperative Diffusion Recovery Likelihood (ICLR 2024 · similarity 0.86)
- Elucidating the design space of classifier-guided diffusion generation (ICLR 2024 · similarity 0.85)
- Efficient Image-to-Image Diffusion Classifier for Adversarial Robustness (AAAI 2025 · similarity 0.85)
- Enhancing Consistency-Based Image Generation via Adversarialy-Trained Classification and Energy-Based Discrimination (NeurIPS 2024 · similarity 0.85)
- Improving Adversarial Energy-Based Model via Diffusion Process (ICML 2024 · similarity 0.85)
- ADDP: Learning General Representations for Image Recognition and Generation with Alternating Denoising Diffusion Process (ICLR 2024 · similarity 0.84)
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.
