可视化看做成理想的猫
加载VGG网络：
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 jsonnp.set_printoptions(precision=3,suppress=True)
model = keras.applications.VGG16(weights='imagenet',include_top=True)
classes = json.loads(open('imagenet_classes.json','r').read())
优化结果：用梯度下降优化，将网络看成猫
绘图：
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]))
结果：
用规格化normalize()围在0-1
代码：
def plot_result(x):
res = model(x)[0]
cls = tf.argmax(res)
print(f"Predicted class: {cls} ({classes[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)
结果；
Predicted class: 669 (mosquito net)
Probability of predicted class = 0.05466596782207489
用VGG调整图像类别概率步骤：
-
选类别，用梯度下降调整图像x（网络V预测类的概率最大化）
-
用交叉熵损失L=L(c,V)，c=目标类别索引
-
梯度下降更新：套公式
-
迭代优化：重复更新epoch，不断调整观察
上图的突出，实际概率可能只有5%（看坐标轴范围定概率），计算用GPU加速，资源有限就减少epoch
target = [284] # Siamese cat
def cross_entropy_loss(target,res):
return tf.reduce_mean(keras.metrics.sparse_categorical_crossentropy(target,res))
def optimize(x,target,epochs=1000,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)
绘图：
Predicted class: 284 (Siamese cat, Siamese) Probability of predicted class = 0.6449142098426819
plot_result(x)
def total_loss(target,res):
return 10*tf.reduce_mean(keras.metrics.sparse_categorical_crossentropy(target,res)) + \
0.005*tf.image.total_variation(x,res)
optimize(x,target,loss_fn=total_loss)
上图需要降噪
减少噪声的方法：用变化损失：衡量相邻像素差
公式（拉乘）：总损失函数=交叉熵损失（分类目标）+λ*变压损失（平滑约束）
步骤：算交叉熵损失 - 算变化损失 - 联合优化（调整相似，总损失最小化）
对平衡系数λ的观察：太小=噪声明显，太大=模糊，分类概率下降；实践中初始值最好0.1-1.0，后面微调，用可视化结果判断最佳平衡点
绘图：
换一个物体：斑马
x = tf.Variable(tf.random.normal((1,224,224,3)))
optimize(x,[340],loss_fn=total_loss) # zebra
对抗攻击：
img = Image.open('images/dog-from-unsplash.jpg')
img = img.crop((200,20,600,420)).resize((224,224))
img = np.array(img)
plt.imshow(img)
绘图：
plot_result(np.expand_dims(img,axis=0))
把狗优化成猫：
x = tf.Variable(np.expand_dims(img,axis=0).astype(np.float32)/255.0)
optimize(x,target,epochs=100)
用ResNet：model作为全局变量，不用VGG
model = keras.applications.ResNet50(weights='imagenet',include_top=True)
绘图：
plot_result(x)
x = tf.Variable(tf.random.normal((1,224,224,3)))optimize(x,target=[340],epochs=500,loss_fn=total_loss)
输出：Epoch: 450, loss: [46.166]
绘图：
plot_result(x)
Predicted class: 340 (zebra)
Probability of predicted class = 0.8876020312309265
用不同的优化器：Keras内置很多优化器可代替梯度下降，将x.assign_sub(eta*grads)换apply)gradients()
def optimize(x,target,epochs=1000,show_every=None,loss_fn=cross_entropy_loss,optimizer=keras.optimizers.SGD(learning_rate=1)):
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
扩展阅读：
-
dropout实验
-
TensorFlow在迁移学习中的运用
-
PyTorch在迁移学习中的运用
-
宠物真实照片分类