Distilling Out-of-Distribution Robustness from Vision-Language Foundation Models
We propose a conceptually simple and lightweight framework for improving the robustness of vision models through the combination of knowledge distillation and data augmentation. We address the conjecture that larger models do not make for better teachers by showing strong gains in out-of-distribution robustness when distilling from pretrained foundation models. Following this finding, we propose Discrete Adversarial Distillation (DAD), which leverages a robust teacher to generate adversarial examples and a VQGAN to discretize them, creating more informative samples than standard data augmentation techniques. We provide a theoretical framework for the use of a robust teacher in the knowledge distillation with data augmentation setting and demonstrate strong gains in out-of-distribution robustness and clean accuracy across different student architectures. Notably, our method adds minor computational overhead compared to similar techniques and can be easily combined with other data augmentations for further improvements.
PDF Abstract NeurIPS 2023 PDF NeurIPS 2023 Abstract
ImageNet
ImageNet-C
ImageNet-R
ImageNet-A
ImageNet-Sketch
Stylized ImageNet
| Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Benchmark |
|---|---|---|---|---|---|---|
| Image Classification | ImageNet | Discrete Adversarial Distillation (ViT-B, 224) | Top 1 Accuracy | 81.9% | # 543 | |
| Domain Generalization | ImageNet-A | Discrete Adversarial Distillation (ResNet-50) | Top-1 accuracy % | 7.7 | # 33 | |
| Domain Generalization | ImageNet-A | Discrete Adversarial Distillation (ViT-B/224) | Top-1 accuracy % | 31.8 | # 26 | |
| Domain Generalization | ImageNet-R | Discrete Adversarial Distillation (ViT-B,224) | Top-1 Error Rate | 34.9 | # 14 | |
| Domain Generalization | ImageNet-Sketch | Discrete Adversarial Distillation (ViT-B, 224) | Top-1 accuracy | 46.1 | # 13 | |
| Image Classification | ImageNet V2 | Discrete Adversarial Distillation (ViT-B, 224) | Top 1 Accuracy | 71.7 | # 22 |
