三、生成heatmap(二）基于patch画热力图
太长不看版
1、将PIL.Image转换成批训练的DataLoader
- 为什么一批一批进去处理
2、载入网络（ torch.load('Resnet.pkl') ），并将数据放入网络，通过 outputs = model(images) 得到预测值，放在对应的对象中
3、按对象的数字顺序排序，生成热力图保存
耐心看完版
事先准备：
test_path = r'C:\Users\BME419\Desktop\resnet\slide\patch'
background_path = r'E:\WSI\CAMELYON16\Processed\patch-based-classification\raw-data\test\tumor091heatmaps patches\none'
pre_savename = r'C:\Users\BME419\Desktop\resnet\slide\heatmap'
savename = os.path.join(pre_savename, 'heatmap') #保存名字为'heatmap'
batch_size = 64
classes = ['negative','positive']
global positive_prob
positive_prob = [] #positive_prob类型为listdef reload_net(model_name): #可选择四种网络
if model_name == "VGG":
trainednet = torch.load('VGGnet.pkl')
elif model_name == "Google":
trainednet = torch.load('Google.pkl')
elif model_name == "Res":
trainednet = torch.load('Resnet.pkl')
elif model_name == "Alex":
trainednet = torch.load('Alexnet.pkl')
return trainednet主函数：
test_data("Res", 224) #内涵调用heatmap_gen()
具体函数 ↓ ↓
def test_data(model_name, input_size): # input_size = 224 # 先转换成 torch 能识别的 Dataset
testset = torchvision.datasets.ImageFolder(test_path,
transform = transforms.Compose([
transforms.Resize((input_size, input_size)),
# 将图片缩放到指定大小（h,w）或者保持长宽比并缩放最短的边到int大小
transforms.ToTensor(),
]))
# 把 dataset 放入 DataLoader
testloader = torch.utils.data.DataLoader(testset, batch_size = batch_size,
shuffle = False, num_workers = 0) # shuffle = False（不打乱）,按顺序取patch，否则 = True,随机取
model = reload_net(model_name) # load模型，函数具体定义见“事先准备”
model.eval() # 把BN和Dropout固定住，不会取平均，而是用训练好的值
产生的Dataset和DataLoader
#将testset.imgs从tuple变为list（用于append网络产生的概率（outputs））
for j in range(len(testset.imgs)):
testset.imgs[j] = list(testset.imgs[j])
#利用训练好的网络预测patch概率
for i, data in enumerate(testloader, 0):
images, labels = data
print(labels)
images = Variable(images, requires_grad=True) # 转换数据格式用Variable
if torch.cuda.is_available():
images = images.cuda() # 转换数据格式用Variable
model.cuda()
with torch.no_grad():
outputs = model(images)
outputs = outputs.cpu().numpy() # 将outputs由GPU转化为numpy
positive_prob.extend(outputs[:, 1]) # positive_prob（list）:各个patch肿瘤positive的概率
if i <= (len(testset.imgs) / 64-1): # 将testset.imgs与各自概率一一对应
for j in range(64):
testset.imgs[j + 64 * i].append(outputs[j, 1])
else:
for j in range(len(testset.imgs) - 64 * i):
testset.imgs[j + 64 * i].append(outputs[j, 1])
probmin = np.min(positive_prob) # 用于背景显示
heatmap_gen(positive_prob, testset,input_size,probmin) #调用heatmap_gen（），具体见↓
0表示'negative', 1表示'positive'
def heatmap_gen(positive_prob,testset,input_size,probmin):
fig = plt.figure(figsize=(172, 153)) #figsize以英寸为单位 width=172，height=153
fig.subplots_adjust(left=0, right=1, bottom=0, top=1, wspace=0, hspace=0)
# 读背景patch
testset2 = torchvision.datasets.ImageFolder(background_path,
transform=transforms.Compose([
transforms.Resize((input_size, input_size)),
# 将图片缩放到指定大小（h,w）或者保持长宽比并缩放最短的边到int大小
transforms.ToTensor(), # 把一个取值范围是[0,255]的PIL.Image 转换成 Tensor
]))
for n in range(len(testset2.imgs)): # 将背景patch概率减10，并append到testset
testset2.imgs[n] = list(testset2.imgs[n])
testset2.imgs[n].append(probmin-10)
testset.imgs.append(testset2.imgs[n])
#使路径减成数字，排序
for n in range(len(testset.imgs)):
testset.imgs[n][0] = os.path.basename(testset.imgs[n][0])
testset.imgs.sort(key=lambda x: int(x[0][:-6]))
positive_prob = [None] * len(testset.imgs)
positive_prob = np.array(positive_prob)
# 按照片名从小到大生成positive_prob
for n in range(len(testset.imgs)):
positive_prob[n] = testset.imgs[n][2]
probmin = np.min(positive_prob)
probmax = np.max(positive_prob)
heatmap = ((positive_prob - probmin) / (probmax - probmin + 0.000001)) * 255 # float在[0，1]之间，转换成0-255
heatmap = heatmap.astype(np.uint8) # 转成unit8
heatmap = heatmap.reshape(153, 172) # y为行，x为列
heatmap = cv2.applyColorMap(heatmap, cv2.COLORMAP_JET) # 生成heat map
heatmap = heatmap[:, :, ::-1] # 注意cv2（BGR）和matplotlib(RGB)通道是相反的
plt.imshow(heatmap)
fig.savefig(savename, dpi=10) #dpi指每英寸有多少个像素，save路径见最上面