Adversarial Cat_TF

import tensorflow as tf
from tensorflow import keras
import matplotlib.pyplot as plt
import numpy as np
from IPython.display import clear_output
from PIL import Image
import json
np.set_printoptions(precision=3,suppress=True)

model = keras.applications.VGG16(weights=None, include_top=True)
classes = [f"fixture-class-{index}" for index in range(1000)]

x = tf.Variable(tf.random.normal((1,224,224,3)))

def normalize(img):
    return (img-tf.reduce_min(img))/(tf.reduce_max(img)-tf.reduce_min(img))

plt.imshow(normalize(x[0]))

def plot_result(x):
    res = model(x)[0]
    cls = tf.argmax(res)
    print(f"Predicted class: {cls} ({classes[int(cls)]})")
    print(f"Probability of predicted class = {res[cls]}")
    fig,ax = plt.subplots(1,2,figsize=(15,2.5),gridspec_kw = { "width_ratios" : [1,5]} )
    ax[0].imshow(normalize(x[0]))
    ax[0].axis('off')
    ax[1].bar(range(1000),res,width=3)
    plt.show()

plot_result(x)

target = [284] # Siamese cat

def cross_entropy_loss(target,res):
    return tf.reduce_mean(keras.losses.sparse_categorical_crossentropy(tf.convert_to_tensor(target, dtype=tf.int32), res))

def optimize(x,target,epochs=10,show_every=None,loss_fn=cross_entropy_loss, eta=1.0):
    if show_every is None:
        show_every = epochs // 10
    for i in range(epochs):
        with tf.GradientTape() as t:
            res = model(x)
            loss = loss_fn(target,res)
            grads = t.gradient(loss,x)
            x.assign_sub(eta*grads)
            if i%show_every == 0:
                clear_output(wait=True)
                print(f"Epoch: {i}, loss: {loss}")
                plt.imshow(normalize(x[0]))
                plt.show()

optimize(x,target)

plot_result(x)

def total_loss(target,res):
    return 10*tf.reduce_mean(keras.losses.sparse_categorical_crossentropy(tf.convert_to_tensor(target, dtype=tf.int32), res)) + \
           0.005*tf.image.total_variation(x)

optimize(x,target,loss_fn=total_loss)

plot_result(x)

x = tf.Variable(tf.random.normal((1,224,224,3)))
optimize(x,[340],loss_fn=total_loss) # zebra

# Generated image fixture replaces the excluded third-party photograph.
yy, xx = np.mgrid[:224, :224]
img = np.stack((
    (xx * 3 + 40) % 256,
    (yy * 2 + 80) % 256,
    ((xx + yy) * 2 + 20) % 256,
), axis=-1).astype(np.uint8)
plt.imshow(img)

plot_result(np.expand_dims(img,axis=0))

x = tf.Variable(np.expand_dims(img,axis=0).astype(np.float32)/255.0)
optimize(x,target,epochs=10)

plot_result(x)

model = keras.applications.ResNet50(weights=None, include_top=True)

plot_result(x)

x = tf.Variable(tf.random.normal((1,224,224,3)))
optimize(x,target=[340],epochs=10,loss_fn=total_loss)

plot_result(x)

def optimize(x,target,epochs=10,show_every=None,loss_fn=cross_entropy_loss,optimizer=keras.optimizers.SGD(learning_rate=1.0)):
    if show_every is None:
        show_every = epochs // 10
    for i in range(epochs):
        with tf.GradientTape() as t:
            res = model(x)
            loss = loss_fn(target,res)
            grads = t.gradient(loss,x)
            optimizer.apply_gradients([(grads,x)])
            if i%show_every == 0:
                clear_output(wait=True)
                print(f"Epoch: {i}, loss: {loss}")
                plt.imshow(normalize(x[0]))
                plt.show()

x = tf.Variable(tf.random.normal((1,224,224,3)))

optimize(x,[898],loss_fn=total_loss) # water bottle