Image Classification on ImageNet Erased Classes
86Church AccuracySD-v1.5
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SD-v1.5Diffusion Model Backbone=SD-v1.5, Classifier Backbone=ResNet-502026.02 | 86 | — | 81 | 82.8 | 94 | 71 | 99 | 97 | 100 | 96 | 99 | 97 | 97 | 92.7 | |
| Neuron-Max Scaling (NMS)Diffusion Model Backbone=SD-v1.5, Classifier Backbone=ResNet-50, Sign Retention Threshold=Top-0.62026.02 | 63 | — | 56 | 31 | 71 | 40 | 26 | 58 | 80 | 69 | 60 | 45 | 48 | 54 | |
| Neuron AverageDiffusion Model Backbone=SD-v1.5, Classifier Backbone=ResNet-50, Sign Retention Threshold=Top-0.62026.02 | 18 | — | 25 | 14 | 13 | 0 | 4 | 3 | 52 | 36 | 6 | 7 | 14 | 16 | |
| Neuron SampleDiffusion Model Backbone=SD-v1.5, Classifier Backbone=ResNet-50, Sign Retention Threshold=Top-0.62026.02 | 17 | — | 26 | 14 | 11 | 1 | 2 | 3 | 49 | 46 | 7 | 8 | 13 | 16 | |
| Concept PruneDiffusion Model Backbone=SD-v1.5, Classifier Backbone=ResNet-502026.02 | 8 | — | 16 | 7 | 6 | 0 | 2 | 1 | 32 | 22 | 2 | 4 | 1 | 8 | |
| Quant RecoverDiffusion Model Backbone=SD-v1.5, Classifier Backbone=ResNet-502026.02 | 7 | — | 19 | 8 | 8 | 2 | 36 | 0 | 31 | 21 | 0 | 3 | 3 | 12 |