test_function.py

import torch
from torch.autograd import Variable


def test(G, f, target, is_targeted, thres, test_loader, epoch, epochs, device, verbose=True):
    n = 0
    acc = 0

    G.eval()
    for i, (img, label) in enumerate(test_loader):
        img_real = Variable(img.to(device))

        pert = torch.clamp(G(img_real), -thres, thres)
        img_fake = pert + img_real
        img_fake = img_fake.clamp(min=0, max=1)

        y_pred = f(img_fake)

        if is_targeted:
            y_target = Variable(torch.ones_like(label).fill_(target).to(device))
            acc += torch.sum(torch.max(y_pred, 1)[1] == y_target).item()
        else:
            y_true = Variable(label.to(device))
            acc += torch.sum(torch.max(y_pred, 1)[1] != y_true).item()

        n += img.size(0)

        if verbose:
            print('Test [%d/%d]: [%d/%d]' %(epoch+1, epochs, i, len(test_loader)), end="\r")
    return acc/n