Diffusion Classifier

Reported on 5 benchmarks across 4 tasks · 2 papers · 2 SOTA

Note: results are matched by exact model name. Different papers may use the same name for different model variants.

Adversarial2 results

Adversarial DefenseonCIFAR-10
Accuracy· 2023-05-24
89.85
best: 90.03 (WRN-28-10)
SOTA
Robust Classification via a Single Diffusion Model arXiv:2305.15241
Adversarial DefenseonCIFAR-10
Robust Accuracy· 2023-05-24
75.67
best: 82.96 (ResNet18 (TRADES-ANCRA/PGD-40))
SOTA
Robust Classification via a Single Diffusion Model arXiv:2305.15241

Computer Vision2 results

Image ClassificationonObjectNet (ImageNet classes)
Top 1 Accuracy· 2023-03-28
33.9
best: 43.4 (Diffusion Classifier (zero-shot))
Your Diffusion Model is Secretly a Zero-Shot Classifier arXiv:2303.16203
Domain GeneralizationonImageNet-A
Top-1 accuracy %· 2023-03-28
30.2
best: 94.17 (Model soups (BASIC-L))
Your Diffusion Model is Secretly a Zero-Shot Classifier arXiv:2303.16203

Methodology1 result

Domain AdaptationonImageNet-A
Top-1 accuracy %· 2023-03-28
30.2
best: 94.17 (Model soups (BASIC-L))
Your Diffusion Model is Secretly a Zero-Shot Classifier arXiv:2303.16203