unbuntu 16.04 MS-Celeb-1M + alexnet + pytorch

最近被保研的事情搞的头大，拖了半天才勉强算结束这个了。从熟悉unbantu 16.04的环境（搭个翻墙的梯子都搞了一上午 呸！）到搭建python,pytorch环境。然后花了一个上午熟悉py的基本语法就开始强撸了，具体的过程等保研结束了再补吧，贴个代码意思一下先。

数据集用的是清洗过的MS-Celeb-1M(em...怎么清洗的之后再补吧)

python用的不熟，踩了很多坑，用pytorch的时候也是，打死不看python中文的官方文档（http://pytorch-cn.readthedocs.io/zh/latest/package_references/functional/ 真香）

后续的慢慢补吧很多细节也在完善....

import torch
import torch.utils.data as data
import torchvision.transforms as transforms
import torchvision.models as models
from torch.autograd import Variable
import torch.optim as optim
import torch.nn as nn
import torch.nn.functional as F
import numpy as np
from torch.utils.data import DataLoader
from PIL import Image
from torch.optim import lr_scheduler
# ------------------ ready for the dataset  ------------------
transform = transforms.Compose([
    transforms.Scale(227),
    transforms.CenterCrop(227),
    transforms.ToTensor(),
    transforms.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225))])

class MyDateset (data.Dataset):

    def __init__(self,data_txt,transform):
        imgs = []
        with open(data_txt , 'r') as f:
            for line in f:
                line = line.strip('\n')
                line = line.rstrip()
                words = line.split()
                labelList = int(words[1])
                imageList = words[0]
                imgs.append((imageList, labelList))

        self.transform = transform
        self.imgs = imgs

    def __getitem__(self, index):
        image_dir,target = self.imgs[index]
        image = Image.open(image_dir)
        image = transform(image)

        return image, target

    def __len__(self):
        return len(self.imgs)

train_data = MyDateset("/home/fuckman/FaceImage/Traindata.txt", transform)
train_loader = DataLoader(train_data,batch_size = 128 ,shuffle=True,num_workers= 8,drop_last=False)

# for img,label in train_data:
 #   print(img.size(),label)

text_data = MyDateset("/home/fuckman/FaceImage/Testdata.txt", transform)
test_loader = DataLoader(dataset=text_data, batch_size = 128 ,shuffle=False, num_workers= 8, drop_last=False)

# print(train_data.__len__())

# --------------- creat the net and train --------------------

class Net(torch.nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = torch.nn.Sequential(
            torch.nn.Conv2d(3, 96, 11, 4, 0),
            torch.nn.ReLU(),
            torch.nn.MaxPool2d(3, 2)
        )
        self.conv2 = torch.nn.Sequential(
            torch.nn.Conv2d(96, 256, 5, 1, 2),
            torch.nn.ReLU(),
            torch.nn.MaxPool2d(3, 2)
        )
        self.conv3 = torch.nn.Sequential(
            torch.nn.Conv2d(256, 384, 3, 1, 1),
            torch.nn.ReLU(),
        )
        self.conv4 = torch.nn.Sequential(
            torch.nn.Conv2d(384, 384, 3, 1, 1),
            torch.nn.ReLU(),
        )
        self.conv5 = torch.nn.Sequential(
            torch.nn.Conv2d(384, 256, 3, 1, 1),
            torch.nn.ReLU(),
            torch.nn.MaxPool2d(3, 2)
        )
        self.dense = torch.nn.Sequential(
            torch.nn.Dropout(0.5),
            torch.nn.Linear(9216, 4096),
            torch.nn.ReLU(),
            torch.nn.Dropout(0.5),
            torch.nn.Linear(4096, 4096),
            torch.nn.ReLU(),
            torch.nn.Linear(4096,1000)
        )

    def forward(self, x):

        conv1_out = self.conv1(x)
        conv2_out = self.conv2(conv1_out)
        conv3_out = self.conv3(conv2_out)
        conv4_out = self.conv4(conv3_out)
        conv5_out = self.conv5(conv4_out)
        res = conv5_out.view(conv5_out.size(0), -1)
        out = self.dense(res)
        return out

alexnet = Net()
alexnet.load_state_dict(torch.load('net_params.pkl'))
alexnet.cuda()

# print( alexnet )

#----------------- training ----------------

# crossentryopyloss
criterion = nn.CrossEntropyLoss()

# SGD with momentum
optimizer = optim.SGD(alexnet.parameters(),lr = 0.01, momentum = 0.9)

# learning rate decay
scheduler = lr_scheduler.MultiStepLR(optimizer, milestones=[10,60], gamma=0.1)

# training
for epoch in range(100):
    scheduler.step()
    running_loss = 0.0
    for i,  data in enumerate(train_loader,0):
        inputs, labels = data
        # print(inputs.size())
        # labels have to be longTensor
        # inputs, labels = Variable(inputs),Variable(labels).long()
        inputs, labels = Variable(inputs.cuda()),Variable(labels.cuda()).long()

        optimizer.zero_grad()
        # inputs should be n * c * w * h n refer to the mini-batch  c refer to the num of channel
        outputs = alexnet(inputs)
        # print(outputs)
        criterion.cuda()

        # outputs should be N*C  labels should be N   N refer to the mini-batch    c refer to the num of class
        # print the size of outputs and labels may help you find the question
        loss = criterion(outputs, labels)
        loss.backward()
        optimizer.step()

        running_loss += loss.data[0]
        if i % 100 == 99:
            print('[%d, %5d] loss : %.3f' %(epoch+1,i+1,running_loss / 100))
            running_loss = 0.0
    if epoch % 10 == 9:
        torch.save(alexnet.state_dict(), 'net_params.pkl')
        print("success")

print("Finished Training")
torch.save(alexnet.state_dict(), 'net_params.pkl')

# ----------------- Test -------------------

correct =0
total = 0
for i, data in enumerate(test_loader,0):
    images, labels = data
    labels = labels.cuda()
    # outputs = alexnet(Variable(images))
    outputs = alexnet(Variable(images.cuda()))
    _, predicted = torch.max(outputs.data, 1)  # max_value and the index of max
    total += labels.size(0)
    new_label = labels.int()
    #print(predicted)
    #print(labels)
    new_predic = predicted.int()
    correct += (new_predic  ==  new_label).sum()

print('Accuracy of the network on the 1000 test images: %d %%' % (100 * correct / total))