first commit

Author: lilei1128

1.jpg

@@ -0,0 +1,123 @@
+import numpy as np
+import torch
+import torchvision
+from torch import nn,optim
+from config import config
+
+
+
+def l2_norm(x):
+    norm = torch.norm(x,p =2 ,dim =1 ,keepdim= True)
+    x = torch.div(x,norm)
+    return x
+# 自己搭建一个简单卷积神经网络
+class myModel(nn.Module):
+    def __init__(self,num_classes):
+        super(myModel,self).__init__()
+        self.layer1 = nn.Sequential(
+            nn.Conv2d(3,16,3), #in_channels out_channels kernel_size
+            nn.BatchNorm2d(16),
+            nn.ReLU(True),
+            nn.MaxPool2d(kernel_size= 2,stride = 2)  #149
+        )
+        self.layer2 = nn.Sequential(
+            nn.Conv2d(16,32,3,2),  #74    #
+            nn.BatchNorm2d(32),
+            nn.ReLU(True),
+            nn.MaxPool2d(kernel_size =2,stride=2)  #37
+
+        )
+        self.layer3 = nn.Sequential(
+            nn.Conv2d(32,32,3,2),  #18
+            nn.BatchNorm2d(32),
+            nn.ReLU(True),
+            nn.MaxPool2d(kernel_size= 2, stride = 2)  #9
+        )
+        self.fc1 = nn.Sequential(
+            nn.Linear(2592,120),
+            nn.ReLU(True)
+        )
+        self.fc2 = nn.Sequential(
+            nn.Linear(120,84),
+            nn.ReLU(True),
+            nn.Linear(84,num_classes)
+        )
+
+    def forward(self, x):
+        x = self.layer1(x)
+        x = self.layer2(x)
+        x = self.layer3(x)
+        x = x.view(x.size(0),-1)
+        x = self.fc1(x)
+        x = self.fc2(x)
+        return x
+
+
+
+class ResNet18(nn.Module):
+    def __init__(self,model,num_classes = 1000):
+        super(ResNet18,self).__init__()
+        self.backbone = model
+
+        self.fc1 = nn.Linear(512,1024)
+        self.dropout = nn.Dropout(0.5)
+        self.fc2 = nn.Linear(1024,num_classes)
+    def forward(self, x):
+        x = self.backbone.conv1(x)
+        x= self.backbone.bn1 (x)
+        x = self.backbone.relu(x)
+        x= self.backbone.maxpool(x)
+
+        x= self.backbone.layer1(x)
+        x = self.backbone.layer2(x)
+        x= self.backbone.layer3(x)
+        x = self.backbone.layer4(x)
+
+        x = self.backbone.avgpool(x)
+
+        x= x.view(x.size(0),-1)
+        x= l2_norm(x)
+        x = self.dropout(x)
+        x = self.fc1(x)
+        x = l2_norm(x)
+        x = self.dropout(x)
+        x = self.fc2(x)
+        return x
+class ResNet101(nn.Module):
+    def __init__(self,model,num_classes =1000):
+        super(ResNet101,self).__init__()
+        self.backbone = model
+
+        self.fc1 = nn.Linear(2048,2048)
+        self.dropout = nn.Dropout(0.5)
+        self.fc2 = nn.Linear(2048,num_classes)
+
+    def forward(self,x):
+        x = self.backbone.conv1(x)
+        x =  self.backbone.bn1(x)
+        x = self.backbone.relu(x)
+        x = self.backbone.maxpool(x)
+
+        x = self.backbone.layer1(x)
+        x = self.backbone.layer2(x)
+        x = self.backbone.layer3(x)
+        x = self.backbone.layer4(x)
+
+        x = self.backbone.avgpool(x)
+
+        x = x.view(x.size(0),-1)
+        x = l2_norm(x)
+        x = self.dropout(x)
+        x = self.fc1(x)
+        x = l2_norm(x)
+        x = self.dropout(x)
+        x = self.fc2(x)
+
+        return x
+
+def get_net():
+    #backbone = torchvision.models.resnet18(pretrained=True)
+    #models = ResNet18(backbone,config.num_classes)
+    backbone = torchvision.models.resnet101(pretrained=True)
+    models = ResNet101(backbone, config.num_classes)
+    return models

2.jpg

@@ -0,0 +1,21 @@
+#可以根据自己的情况进行修改
+class MyConfigs():
+
+    data_folder = './data/flowers/'
+    test_data_folder = ""
+    model_name = "resnet" #Vgg ResNet152 myModel
+    weights = "./checkpoints/"
+    logs = "./logs/"
+    example_folder = "./example/"
+    freeze = True
+    #
+    epochs = 300
+    batch_size = 16
+    img_height = 227  #网络输入的高和宽
+    img_width = 227
+    num_classes = 20
+    lr = 1e-2
+    lr_decay = 1e-4
+    weight_decay = 2e-4
+    ratio = 0.2
+config = MyConfigs()

Model.py

@@ -0,0 +1,171 @@
+import torch
+from torch.utils.data import Dataset,DataLoader
+import cv2
+import os
+from tqdm import tqdm
+from config import config
+from glob import glob
+import os
+from torchvision import transforms
+import numpy as np
+import random
+from shutil import copy
+from PIL import Image
+import math
+
+np.random.seed(666) #设置随机种子 为了保证每次划分训练集和测试机的是相同的
+
+
+'''
+# 1. 对于mini_data　数据集的解析
+def parse_data_config(data_path):
+    files = []
+    #
+    for img in os.listdir(data_path):
+        image = data_path + img
+        label = img.split("__")[0][3:]
+        files.append((image,label))
+    return files
+
+#划分训练集和测试集
+# ratio 为划分为测试集的比例
+def divide_data(data_path,ratio):
+    files = parse_data_config(data_path)
+    temp = np.array(files)
+    test_data = []
+    train_data = []
+    for i in range(config.num_classes):
+        temp_data = []
+        for data in temp:
+            if data[1] == str(i):
+                temp_data.append(data)
+        np.random.shuffle(np.array(temp_data))
+        test_data =test_data + temp_data[:int(ratio * len(temp_data))]
+        train_data = train_data + temp_data[int(ratio*len(temp_data))+1:]
+    # np.random.shuffle(temp)
+    # test_data = files[:int(ratio * len(files))]
+    # train_data = files[int(ratio*len(files))+1:]
+
+    # 从训练集中挑选 10 中图片保存到 example 文件夹中
+    if not os.path.exists(config.example_folder):
+        os.mkdir(config.example_folder)
+    else:
+        for i in os.listdir(config.example_folder):
+            os.remove(os.path.join(config.example_folder+i))
+    for i in range(10):
+        index = random.randint(0,len(test_data)-1)  # 随机生成图片的索引
+        copy(test_data[index][0],config.example_folder)  # 将挑选的图像复制到example文件夹
+
+    return test_data, train_data
+'''
+# 2. 对于flowers 数据集的解析
+def get_files(file_dir,ratio):
+    roses = []
+    labels_roses = []
+    tulips = []
+    labels_tulips = []
+    dandelion = []
+    labels_dandelion=[]
+    sunflowers = []
+    labels_sunflowers = []
+    for file  in os.listdir(file_dir +'roses'):
+        roses.append(file_dir + 'roses' + '/' + file)
+        labels_roses.append(0)
+    for file in os.listdir(file_dir + 'tulips'):
+        tulips.append(file_dir + 'tulips' + '/' + file)
+        labels_tulips.append(1)
+    for file in os.listdir(file_dir + 'dandelion'):
+        tulips.append(file_dir + 'dandelion' + '/' +file)
+        labels_dandelion.append(2)
+    for file in os.listdir(file_dir + 'sunflowers'):
+        sunflowers.append(file_dir + 'sunflowers' + '/' +file)
+        labels_sunflowers.append(3)
+
+    image_list = np.hstack((roses ,tulips, dandelion, sunflowers))
+    labels_list = np.hstack((labels_roses, labels_tulips, labels_dandelion, labels_sunflowers))
+    temp = np.array([image_list, labels_list])
+    temp = temp.transpose()
+    np.random.shuffle(temp)
+    all_image_list = list(temp[:,0])
+    all_label_list = list(temp[:,1])
+    all_label_list = [int(i) for i in all_label_list]
+    length = len(all_image_list)
+    n_test = int(math.ceil(length * ratio))
+    n_train = length - n_test
+
+    tra_image = all_image_list[0:n_train]
+    tra_label = all_label_list[0:n_train]
+
+    test_image = all_image_list[n_train:-1]
+    test_label = all_label_list[n_train:-1]
+
+    train_data = [(tra_image[i],tra_label[i]) for i in range(len(tra_image))]
+    test_data = [(test_image[i],test_label[i]) for i in range(len(test_image))]
+   # print("train_data = ",test_image)
+   # print("test_data = " , test_label)
+    return test_data,train_data
+
+#这个数据集类的作用就是加载训练和测试时的数据
+class datasets(Dataset):
+    def __init__(self,data,transform = None,test = False):
+        imgs = []
+        labels = []
+        self.test = test
+        self.len = len(data)
+        self.data = data
+        self.transform = transform
+        for i in self.data:
+            imgs.append(i[0])
+            self.imgs = imgs
+            labels.append(int(i[1]) ) #pytorch中交叉熵需要从0开始
+            self.labels  = labels
+    def __getitem__(self,index):
+        if self.test:
+            filename = self.imgs[index]
+            filename = filename
+            img_path = self.imgs[index]
+            img = cv2.imread(img_path)
+            img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
+            img = cv2.resize(img, (config.img_width, config.img_height))
+            img = transforms.ToTensor()(img)
+            return img,filename
+        else:
+            img_path = self.imgs[index]
+            label = self.labels[index]
+            #label = int(label)
+            img = cv2.imread(img_path)
+            img = cv2.cvtColor(img,cv2.COLOR_BGR2RGB)
+            img = cv2.resize(img,(config.img_width,config.img_height))
+            # img = transforms.ToTensor()(img)
+
+            if self.transform is not None:
+                img = Image.fromarray(img)
+                img = self.transform(img)
+
+            else:
+                img = transforms.ToTensor()(img)
+            return img,label
+
+    def __len__(self):
+        return len(self.data)#self.len
+
+def collate_fn(batch): #表示如何将多个样本拼接成一个batch
+    imgs = []
+    label = []
+    for sample in batch:
+        imgs.append(sample[0])
+        label.append(sample[1])
+
+    return torch.stack(imgs, 0),label
+
+
+#用于调试代码
+if __name__ == '__main__':
+    test_data,_ = get_files(config.data_folder,0.2)
+    for i in (test_data):
+        print(i)
+    print(len(test_data))
+
+    transform = transforms.Compose([transforms.ToTensor()])
+    data = datasets(test_data,transform = transform)
+    #print(data[0])