使用VGG模型進行貓狗大戰(zhàn)
import numpy as np import matplotlib.pyplot as plt import os import torch import torch.nn as nn import torchvision from torchvision import models,transforms,datasets import time import json
1、下載數據
! wget https://static.leiphone.com/cat_dog.rar
! unrar x cat_dog.rar
2、數據處理
datasets 是 torchvision 中的一個包，可以用做加載圖像數據。它可以以多線程（multi-thread）的形式從硬盤中讀取數據，使用 mini-batch 的形式，在網絡訓練中向 GPU 輸送。在使用CNN處理圖像時，需要進行預處理。圖片將被整理成 224×224×3 的大小，同時還將進行歸一化處理。
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) vgg_format = transforms.Compose([ transforms.CenterCrop(224), transforms.ToTensor(), normalize, ]) #這里進行了修改，包括訓練數據、驗證數據、以及測試數據，分別在三個目錄train/val/test import shutil data_dir = './cat_dog' os.mkdir("./cat_dog/train/cat") os.mkdir("./cat_dog/train/dog") os.mkdir("./cat_dog/val/cat") os.mkdir("./cat_dog/val/dog") for i in range(10000): cat_name = './cat_dog/train/cat_'+str(i)+'.jpg'; dog_name = './cat_dog/train/dog_'+str(i)+'.jpg'; shutil.move(cat_name,"./cat_dog/train/cat") shutil.move(dog_name,"./cat_dog/train/dog") for i in range(1000): cat_name = './cat_dog/val/cat_'+str(i)+'.jpg'; dog_name = './cat_dog/val/dog_'+str(i)+'.jpg'; shutil.move(cat_name,"./cat_dog/val/cat") shutil.move(dog_name,"./cat_dog/val/dog") #讀取測試問題的數據集 test_path = "./cat_dog/test/dogs_cats" os.mkdir(test_path) #移動到test_path for i in range(2000): name = './cat_dog/test/'+str(i)+'.jpg' shutil.move(name,"./cat_dog/test/dogs_cats") file_list=os.listdir("./cat_dog/test/dogs_cats") #將圖片名補全，防止讀取順序不對 for file in file_list: #填充0后名字總共10位，包括擴展名 filename = file.zfill(10) new_name =''.join(filename) os.rename(test_path+'/'+file,test_path+'/'+new_name) #將所有圖片數據放到dsets內 dsets = {x: datasets.ImageFolder(os.path.join(data_dir, x), vgg_format) for x in ['train','val','test']} dset_sizes = {x: len(dsets[x]) for x in ['train','val','test']} dset_classes = dsets['train'].classes
loader_train = torch.utils.data.DataLoader(dsets['train'], batch_size=64, shuffle=True, num_workers=6) loader_valid = torch.utils.data.DataLoader(dsets['val'], batch_size=5, shuffle=False, num_workers=6) #加入測試集 loader_test = torch.utils.data.DataLoader(dsets['test'], batch_size=5,shuffle=False, num_workers=6) ''' valid 數據一共有2000張圖，每個batch是5張，因此，下面進行遍歷一共會輸出到 400 同時，把第一個 batch 保存到 inputs_try, labels_try，分別查看 ''' count = 1 for data in loader_test: print(count, end=',') if count%50==0: print() if count == 1: inputs_try,labels_try = data count +=1 print(labels_try) print(inputs_try.shape)
# 顯示圖片的小程序 def imshow(inp, title=None): # Imshow for Tensor. inp = inp.numpy().transpose((1, 2, 0)) mean = np.array([0.485, 0.456, 0.406]) std = np.array([0.229, 0.224, 0.225]) inp = np.clip(std * inp + mean, 0,1) plt.imshow(inp) if title is not None: plt.title(title) plt.pause(0.001) # pause a bit so that plots are updated
# 顯示 labels_try 的5張圖片，即valid里第一個batch的5張圖片 out = torchvision.utils.make_grid(inputs_try) imshow(out, title=[dset_classes[x] for x in labels_try])
3. 創(chuàng)建 VGG Model
!wget https://s3.amazonaws.com/deep-learning-models/image-models/imagenet_class_index.json
model_vgg = models.vgg16(pretrained=True) with open('./imagenet_class_index.json') as f: class_dict = json.load(f) dic_imagenet = [class_dict[str(i)][1] for i in range(len(class_dict))] inputs_try , labels_try = inputs_try.to(device), labels_try.to(device) model_vgg = model_vgg.to(device) outputs_try = model_vgg(inputs_try) print(outputs_try) print(outputs_try.shape) ''' 可以看到結果為5行，1000列的數據，每一列代表對每一種目標識別的結果。 但是我也可以觀察到，結果非常奇葩，有負數，有正數， 為了將VGG網絡輸出的結果轉化為對每一類的預測概率，我們把結果輸入到 Softmax 函數 ''' m_softm = nn.Softmax(dim=1) probs = m_softm(outputs_try) vals_try,pred_try = torch.max(probs,dim=1) print( 'prob sum: ', torch.sum(probs,1)) print( 'vals_try: ', vals_try) print( 'pred_try: ', pred_try) print([dic_imagenet[i] for i in pred_try.data]) imshow(torchvision.utils.make_grid(inputs_try.data.cpu()), title=[dset_classes[x] for x in labels_try.data.cpu()])
4. 修改最后一層，凍結前面層的參數
print(model_vgg) model_vgg_new = model_vgg; for param in model_vgg_new.parameters(): param.requires_grad = False model_vgg_new.classifier._modules['6'] = nn.Linear(4096, 2) model_vgg_new.classifier._modules['7'] = torch.nn.LogSoftmax(dim = 1) model_vgg_new = model_vgg_new.to(device) print(model_vgg_new.classifier)
5. 訓練并測試全連接層
包括三個步驟：第1步，創(chuàng)建損失函數和優(yōu)化器；第2步，訓練模型；第3步，測試模型。
''' 第一步：創(chuàng)建損失函數和優(yōu)化器 損失函數 NLLLoss() 的 輸入 是一個對數概率向量和一個目標標簽. 它不會為我們計算對數概率，適合最后一層是log_softmax()的網絡. ''' criterion = nn.NLLLoss() # 學習率 lr = 0.001 # 隨機梯度下降 optimizer_vgg = torch.optim.SGD(model_vgg_new.classifier[6].parameters(),lr = lr) ''' 第二步：訓練模型 ''' def train_model(model,dataloader,size,epochs=1,optimizer=None): model.train() for epoch in range(epochs): running_loss = 0.0 running_corrects = 0 count = 0 for inputs,classes in dataloader: inputs = inputs.to(device) classes = classes.to(device) outputs = model(inputs) loss = criterion(outputs,classes) optimizer = optimizer optimizer.zero_grad() loss.backward() optimizer.step() _,preds = torch.max(outputs.data,1) # statistics running_loss += loss.data.item() running_corrects += torch.sum(preds == classes.data) count += len(inputs) print('Training: No. ', count, ' process ... total: ', size) epoch_loss = running_loss / size epoch_acc = running_corrects.data.item() / size print('Loss: {:.4f} Acc: {:.4f}'.format( epoch_loss, epoch_acc)) # 模型訓練 train_model(model_vgg_new,loader_train,size=dset_sizes['train'], epochs=1, optimizer=optimizer_vgg)
#驗證模型正確率的代碼 def test_model(model,dataloader,size): model.eval() predictions = np.zeros(size) all_classes = np.zeros(size) all_proba = np.zeros((size,2)) i = 0 running_loss = 0.0 running_corrects = 0 for inputs,classes in dataloader: inputs = inputs.to(device) classes = classes.to(device) outputs = model(inputs) loss = criterion(outputs,classes) _,preds = torch.max(outputs.data,1) # statistics running_loss += loss.data.item() running_corrects += torch.sum(preds == classes.data) predictions[i:i+len(classes)] = preds.to('cpu').numpy() all_classes[i:i+len(classes)] = classes.to('cpu').numpy() all_proba[i:i+len(classes),:] = outputs.data.to('cpu').numpy() i += len(classes) print('validing: No. ', i, ' process ... total: ', size) epoch_loss = running_loss / size epoch_acc = running_corrects.data.item() / size print('Loss: {:.4f} Acc: {:.4f}'.format( epoch_loss, epoch_acc)) return predictions, all_proba, all_classes #predictions, all_proba, all_classes = test_model(model_vgg_new,loader_valid,size=dset_sizes['val']) #如果使用的是已有的模型，應該跑下面這行代碼 predictions, all_proba, all_classes = test_model(model_new,loader_valid,size=dset_sizes['val'])
#這個是對測試集進行預測的代碼 def result_model(model,dataloader,size): model.eval() predictions=np.zeros((size,2),dtype='int') i = 0 for inputs,classes in dataloader: inputs = inputs.to(device) outputs = model(inputs) #_表示的就是具體的value，preds表示下標，1表示在行上操作取最大值，返回類別 _,preds = torch.max(outputs.data,1) predictions[i:i+len(classes),1] = preds.to('cpu').numpy(); predictions[i:i+len(classes),0] = np.linspace(i,i+len(classes)-1,len(classes)) #可在過程中看到部分結果 print(predictions[i:i+len(classes),:]) i += len(classes) print('creating: No. ', i, ' process ... total: ', size) return predictions result = result_model(model_vgg_new,loader_test,size=dset_sizes['test']) #如果使用的是已有的模型，應該跑下面這行代碼 result = result_model(model_new,loader_test,size=dset_sizes['test']) #這里是生成結果的文件，上傳到AI研習社可以看到正確率 np.savetxt("./cat_dog/result.csv",result,fmt="%d",delimiter=",")
6. 可視化模型預測結果（主觀分析）
主觀分析就是把預測的結果和相對應的測試圖像輸出出來看看，一般有四種方式：
隨機查看一些預測正確的圖片
隨機查看一些預測錯誤的圖片
預測正確，同時具有較大的probability的圖片
預測錯誤，同時具有較大的probability的圖片
最不確定的圖片，比如說預測概率接近0.5的圖片
# 單次可視化顯示的圖片個數 n_view = 8 correct = np.where(predictions==all_classes)[0] from numpy.random import random, permutation idx = permutation(correct)[:n_view] print('random correct idx: ', idx) loader_correct = torch.utils.data.DataLoader([dsets['valid'][x] for x in idx], batch_size = n_view,shuffle=True) for data in loader_correct: inputs_cor,labels_cor = data # Make a grid from batch out = torchvision.utils.make_grid(inputs_cor) imshow(out, title=[l.item() for l in labels_cor]) print(all_classes) # 類似的思路，可以顯示錯誤分類的圖片，這里不再重復代碼
浙公網安備 33010602011771號