PyramidAT (PyTorch version)

(Original) Paper

Pyramid Adversarial Training Improves ViT's Performance

Official repository with TF/JAX

link

Usage

All you have to do is loading images and deep model(s).

import torchvision
import cv2

from pyramidAT import pyramidAT

n_steps = 10

models = torchvision.models.resnet50(pretrained='imagenet').eval()
images = cv2.resize(cv2.imread('imgs/golf_ball.jfif'), (224,224))

perturbed_image = pyramidAT(images, model, mode='nearest', n_steps=n_steps)

Milestone

• Convert original TF/JAX to PyTorch version
• PyramidAT with dataloader
• Bench-marking table

Contact (or collaborate)

email: kdhht5022@gmail.com