attack_methods.py

import torch
import numpy as np
import scipy.stats as st
import torch.nn.functional as F

"""Translation-Invariant https://arxiv.org/abs/1904.02884"""
def gkern(kernlen=15, nsig=3):
    x = np.linspace(-nsig, nsig, kernlen)
    kern1d = st.norm.pdf(x)
    kernel_raw = np.outer(kern1d, kern1d)
    kernel = kernel_raw / kernel_raw.sum()
    kernel = kernel.astype(np.float32)
    gaussian_kernel = np.stack([kernel, kernel, kernel])  # 5*5*3
    gaussian_kernel = np.expand_dims(gaussian_kernel, 1)  # 1*5*5*3
    gaussian_kernel = torch.from_numpy(gaussian_kernel).cuda()  # tensor and cuda
    return gaussian_kernel

"""Input diversity: https://arxiv.org/abs/1803.06978"""
def DI(x, resize_rate=1.15, diversity_prob=0.5):
    assert resize_rate >= 1.0
    assert diversity_prob >= 0.0 and diversity_prob <= 1.0
    img_size = x.shape[-1]
    img_resize = int(img_size * resize_rate)
    rnd = torch.randint(low=img_size, high=img_resize, size=(1,), dtype=torch.int32)
    rescaled = F.interpolate(x, size=[rnd, rnd], mode='bilinear', align_corners=False)
    h_rem = img_resize - rnd
    w_rem = img_resize - rnd
    pad_top = torch.randint(low=0, high=h_rem.item(), size=(1,), dtype=torch.int32)
    pad_bottom = h_rem - pad_top
    pad_left = torch.randint(low=0, high=w_rem.item(), size=(1,), dtype=torch.int32)
    pad_right = w_rem - pad_left
    padded = F.pad(rescaled, [pad_left.item(), pad_right.item(), pad_top.item(), pad_bottom.item()], value=0)
    ret = padded if torch.rand(1) < diversity_prob else x
    return ret