DiffAug: A Diffuse-and-Denoise Augmentation for Training Robust Classifiers
Fuente:
arXiv
Guardado en:
| Autores principales: | , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2023
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866929362823020544 |
|---|---|
| author | Sastry, Chandramouli Dumpala, Sri Harsha Oore, Sageev |
| author_facet | Sastry, Chandramouli Dumpala, Sri Harsha Oore, Sageev |
| contents | We introduce DiffAug, a simple and efficient diffusion-based augmentation technique to train image classifiers for the crucial yet challenging goal of improved classifier robustness. Applying DiffAug to a given example consists of one forward-diffusion step followed by one reverse-diffusion step. Using both ResNet-50 and Vision Transformer architectures, we comprehensively evaluate classifiers trained with DiffAug and demonstrate the surprising effectiveness of single-step reverse diffusion in improving robustness to covariate shifts, certified adversarial accuracy and out of distribution detection. When we combine DiffAug with other augmentations such as AugMix and DeepAugment we demonstrate further improved robustness. Finally, building on this approach, we also improve classifier-guided diffusion wherein we observe improvements in: (i) classifier-generalization, (ii) gradient quality (i.e., improved perceptual alignment) and (iii) image generation performance. We thus introduce a computationally efficient technique for training with improved robustness that does not require any additional data, and effectively complements existing augmentation approaches. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2306_09192 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | DiffAug: A Diffuse-and-Denoise Augmentation for Training Robust Classifiers Sastry, Chandramouli Dumpala, Sri Harsha Oore, Sageev Computer Vision and Pattern Recognition Machine Learning We introduce DiffAug, a simple and efficient diffusion-based augmentation technique to train image classifiers for the crucial yet challenging goal of improved classifier robustness. Applying DiffAug to a given example consists of one forward-diffusion step followed by one reverse-diffusion step. Using both ResNet-50 and Vision Transformer architectures, we comprehensively evaluate classifiers trained with DiffAug and demonstrate the surprising effectiveness of single-step reverse diffusion in improving robustness to covariate shifts, certified adversarial accuracy and out of distribution detection. When we combine DiffAug with other augmentations such as AugMix and DeepAugment we demonstrate further improved robustness. Finally, building on this approach, we also improve classifier-guided diffusion wherein we observe improvements in: (i) classifier-generalization, (ii) gradient quality (i.e., improved perceptual alignment) and (iii) image generation performance. We thus introduce a computationally efficient technique for training with improved robustness that does not require any additional data, and effectively complements existing augmentation approaches. |
| title | DiffAug: A Diffuse-and-Denoise Augmentation for Training Robust Classifiers |
| topic | Computer Vision and Pattern Recognition Machine Learning |
| url | https://arxiv.org/abs/2306.09192 |