Fixup_Resnet实现CIFAR-10分类精度达到92.29%

最新推荐文章于 2023-05-24 15:37:22 发布

Fu_Xingwen

最新推荐文章于 2023-05-24 15:37:22 发布

阅读量710

点赞数

文章标签： pytorch 神经网络机器学习深度学习

本文链接：https://blog.csdn.net/qq_41249412/article/details/115610251

版权

时间20210407
作者：知道许多的橘子
实现：Fixup_Resnet32对CIFAR10数据集的分类
测试集准确度：92.29%
实现框架pytorch
数据增强方法：FMiX+RandomCrop+RandomHorizontalFlip+Normalize
训练次数：200
阶段学习率[0-100]：0.1||[100-150]：0.01||[150-200]：0.001
优化器：torch.optim.SGD(model.parameters(),lr=learning_rate,momentum=0.9,weight_decay=1e-5)

如果感觉算力不够用了，或者心疼自己电脑了！

可以用我实验室的算力，试试呢！

害，谁叫我的算力都用不完呢！

支持所有框架！实际上框架都配置好了！

傻瓜式云计算！

Tesla v100 1卡，2卡，4卡，8卡

内存16-128G

cpu：8-24核

想要？加个微信：15615634293

或者点这里，找我！

欢迎打扰！

# In[1] 导入所需工具包
import os
import torch
import torch.nn as nn
import torchvision
from torchvision import datasets,transforms
import time
from torch.nn import functional as F
from math import floor, ceil
import math
import numpy as np
from workspace.FMIX.fmix import sample_and_apply, sample_mask
#import torchvision.transforms as transforms
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print(device)
import random
# In[1] 设置超参数
num_epochs = 200
batch_size = 128
tbatch_size = 5000
learning_rate = 0.1
test_name = '/_C10_Fixup_resnet'
class RandomErasing(object):
    '''
    Class that performs Random Erasing in Random Erasing Data Augmentation by Zhong et al.
    -------------------------------------------------------------------------------------
    probability: The probability that the operation will be performed.
    sl: min erasing area
    sh: max erasing area
    r1: min aspect ratio
    mean: erasing value
    -------------------------------------------------------------------------------------
    '''
    def __init__(self, probability = 0.5, sl = 0.02, sh = 0.4, r1 = 0.3, mean=[0.4914, 0.4822, 0.4465]):
        self.probability = probability
        self.mean = mean
        self.sl = sl
        self.sh = sh
        self.r1 = r1

    def __call__(self, img):

        if random.uniform(0, 1) > self.probability:
            return img

        for attempt in range(100):
            area = img.size()[1] * img.size()[2]

            target_area = random.uniform(self.sl, self.sh) * area
            aspect_ratio = random.uniform(self.r1, 1/self.r1)

            h = int(round(math.sqrt(target_area * aspect_ratio)))
            w = int(round(math.sqrt(target_area / aspect_ratio)))

            if w <= img.size()[2] and h <= img.size()[1]:
                x1 = random.randint(0, img.size()[1] - h)
                y1 = random.randint(0, img.size()[2] - w)
                if img.size()[0] == 3:
                    img[0, x1:x1+h, y1:y1+w] = self.mean[0]
                    img[1, x1:x1+h, y1:y1+w] = self.mean[1]
                    img[2, x1:x1+h, y1:y1+w] = self.mean[2]
                else:
                    img[0, x1:x1+h, y1:y1+w] = self.mean[0]
                return img

        return img

# In[1] 加载数据
# In[2]#图像预处理变换的定义
transform_train = transforms.Compose([
transforms.RandomCrop(32, padding=4), #在一个随机的位置进行裁剪，32正方形裁剪，每个边框上填充4
transforms.RandomHorizontalFlip(), #以给定的概率随机水平翻转给定的PIL图像，默认值为0.5
transforms.ToTensor(), #将PIL Image或者 ndarray 转换为tensor，并且归一化至[0-1]
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),#用平均值和标准偏差归一化张量图像，
#(M1,…,Mn)和(S1,…,Sn)将标准化输入的每个通道
])
transform_test = transforms.Compose([ #测试集同样进行图像预处理
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
cifar10_train = torchvision.datasets.CIFAR10(root='/home/megstudio/dataset', train=True, download=False, transform=transform_train)
cifar10_test = torchvision.datasets.CIFAR10(root='/home/megstudio/dataset', train=False, download=False, transform=transform_test)

train_loader = torch.utils.data.DataLoader(cifar10_train, batch_size=batch_size, shuffle=True, num_workers=2)
test_loader = torch.utils.data.DataLoader(cifar10_test, batch_size=tbatch_size, shuffle=False, num_workers=2)

def conv3x3(in_planes, out_planes, stride=1):
    """3x3 convolution with padding"""
    return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride,
                     padding=1, bias=False)


class FixupBasicBlock(nn.Module):
    expansion = 1

    def __init__(self, inplanes, planes, stride=1, downsample=None):
        super(FixupBasicBlock, self).__init__()
        # Both self.conv1 and self.downsample layers downsample the input when stride != 1
        self.bias1a = nn.Parameter(torch.zeros(1))
        self.conv1 = conv3x3(inplanes, planes, stride)
        self.bias1b = nn.Parameter(torch.zeros(1))
        self.relu = nn.ReLU(inplace=True)
        self.bias2a = nn.Parameter(torch.zeros(1))
        self.conv2 = conv3x3(planes, planes)
        self.scale = nn.Parameter(torch.ones(1))
        self.bias2b = nn.Parameter(torch.zeros(1))
        self.downsample = downsample

    def forward(self, x):
        identity = x

        out = self.conv1(x + self.bias1a)
        out = self.relu(out + self.bias1b)

        out = self.conv2(out + self.bias2a)
        out = out * self.scale + self.bias2b

        if self.downsample is not None:
            identity = self.downsample(x + self.bias1a)
            identity = torch.cat((identity, torch.zeros_like(identity)), 1)

        out += identity
        out = self.relu(out)

        return out


class FixupResNet(nn.Module):

    def __init__(self, block, layers, num_classes=10):
        super(FixupResNet, self).__init__()
        self.num_layers = sum(layers)
        self.inplanes = 16
        self.conv1 = conv3x3(3, 16)
        self.bias1 = nn.Parameter(torch.zeros(1))
        self.relu = nn.ReLU(inplace=True)
        self.layer1 = self._make_layer(block, 16, layers[0])
        self.layer2 = self._make_layer(block, 32, layers[1], stride=2)
        self.layer3 = self._make_layer(block, 64, layers[2], stride=2)
        self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
        self.bias2 = nn.Parameter(torch.zeros(1))
        self.fc = nn.Linear(64, num_classes)

        for m in self.modules():
            if isinstance(m, FixupBasicBlock):
                nn.init.normal_(m.conv1.weight, mean=0, std=np.sqrt(2 / (m.conv1.weight.shape[0] * np.prod(m.conv1.weight.shape[2:]))) * self.num_layers ** (-0.5))
                nn.init.constant_(m.conv2.weight, 0)
            elif isinstance(m, nn.Linear):
                nn.init.constant_(m.weight, 0)
                nn.init.constant_(m.bias, 0)

    def _make_layer(self, block, planes, blocks, stride=1):
        downsample = None
        if stride != 1:
            downsample = nn.AvgPool2d(1, stride=stride)

        layers = []
        layers.append(block(self.inplanes, planes, stride, downsample))
        self.inplanes = planes
        for _ in range(1, blocks):
            layers.append(block(planes, planes))

        return nn.Sequential(*layers)

    def forward(self, x):
        x = self.conv1(x)
        x = self.relu(x + self.bias1)

        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)

        x = self.avgpool(x)
        x = x.view(x.size(0), -1)
        x = self.fc(x + self.bias2)

        return x


def fixup_resnet20(**kwargs):
    """Constructs a Fixup-ResNet-20 model.

    """
    model = FixupResNet(FixupBasicBlock, [3, 3, 3], **kwargs)
    return model


def fixup_resnet32(**kwargs):
    """Constructs a Fixup-ResNet-32 model.

    """
    model = FixupResNet(FixupBasicBlock, [5, 5, 5], **kwargs)
    return model


def fixup_resnet44(**kwargs):
    """Constructs a Fixup-ResNet-44 model.

    """
    model = FixupResNet(FixupBasicBlock, [7, 7, 7], **kwargs)
    return model


def fixup_resnet56(**kwargs):
    """Constructs a Fixup-ResNet-56 model.

    """
    model = FixupResNet(FixupBasicBlock, [9, 9, 9], **kwargs)
    return model


def fixup_resnet110(**kwargs):
    """Constructs a Fixup-ResNet-110 model.

    """
    model = FixupResNet(FixupBasicBlock, [18, 18, 18], **kwargs)
    return model


def fixup_resnet1202(**kwargs):
    """Constructs a Fixup-ResNet-1202 model.

    """
    model = FixupResNet(FixupBasicBlock, [200, 200, 200], **kwargs)
    return model    

# In[1] 设置一个通过优化器更新学习率的函数
def update_lr(optimizer, lr):
    for param_group in optimizer.param_groups:
        param_group['lr'] = lr
# In[1] 定义测试函数
def test(model,test_loader):
    
    model.eval()
    with torch.no_grad():
        correct = 0
        total = 0
        for images, labels in test_loader:
            images = images.to(device)
            labels = labels.to(device)
            outputs= model(images)

            _, predicted = torch.max(outputs.data, 1)

            total += labels.size(0)
            correct += (predicted == labels).sum().item()
        acc=100 * correct / total
    
        print('Accuracy of the model on the test images: {} %'.format(acc))
    return acc

# In[1] 定义模型和损失函数
# =============================================================================
def mkdir(path):
    folder = os.path.exists(path)
    if not folder:                   #判断是否存在文件夹如果不存在则创建为文件夹
        os.makedirs(path)            #makedirs 创建文件时如果路径不存在会创建这个路径
        print("---  new folder...  ---")
        print("---  OK  ---")
    else:
        print("---  There is this folder!  ---")
path = os.getcwd()
path = path+test_name
print(path)
mkdir(path)             #调用函数
# In[1] 定义模型和损失函数
#$$
# =============================================================================
try:
    model = torch.load(path+'/model.pkl').to(device)
    learning_rate = np.load(path+'/learning_rate.npy')
    #learning_rate *= 3
    print(learning_rate)
    train_loss = np.load(path+'/test_acc.npy').tolist()
    test_acc = np.load(path+'/test_acc.npy').tolist()
    print("---  There is a model in the folder...  ---")
except:
    print("---  Create a new model...  ---")
    model = fixup_resnet32().to(device)
    train_loss=[]#准备放误差     
    test_acc=[]#准备放测试准确率
# =============================================================================

criterion = nn.CrossEntropyLoss()
#optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate, weight_decay=1e-4)
optimizer = torch.optim.SGD(model.parameters(),lr=learning_rate,momentum=0.9,weight_decay=5e-4) 
scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer,  milestones = [100, 150], gamma = 0.1, last_epoch=-1)
# In[1] 训练模型更新学习率
total_step = len(train_loader)
curr_lr = learning_rate
for epoch in range(num_epochs):
    scheduler.step()
    in_epoch = time.time()
    for i, (images, labels) in enumerate(train_loader):
        
        # =============================================================================
        #FMiX
        #print(type(images))
        #images, index, lam = sample_and_apply(images, alpha=1, decay_power=3, shape=(32,32))
        #images = images.type(torch.FloatTensor) 
        #shuffled_label = labels[index].to(device)
        
        images = images.to(device)
        labels = labels.to(device)
        # Forward pass
        outputs= model(images)
        #loss = lam*criterion(outputs, labels) + (1-lam)*criterion(outputs, shuffled_label)
        loss = criterion(outputs, labels)
        # =============================================================================

        # Backward and optimize
        optimizer.zero_grad()
        loss.backward(retain_graph=False)
        optimizer.step()
        
        if (i + 1) % 100 == 0:
            print("Epoch [{}/{}], Step [{}/{}] Loss: {:.4f}"
                  .format(epoch + 1, num_epochs, i + 1, total_step, loss.item()))
    train_loss.append(loss.item())
    acctemp=test(model,test_loader)
    test_acc.append(acctemp)
    out_epoch = time.time()
    print(f"use {(out_epoch-in_epoch)//60}min{(out_epoch-in_epoch)%60}s")
    #if (epoch + 1) % 30 == 0:
        #curr_lr /= 3
        #update_lr(optimizer, curr_lr)

#$$
# In[1] 测试模型并保存
test(model,train_loader)
torch.save(model, path+'/model.pkl')
# torch.save(model.state_dict(), 'resnet.ckpt')
curr_lr_save=np.array(curr_lr)
np.save(path+'/learning_rate.npy',curr_lr_save)
test_acc=np.array(test_acc)
np.save(path+'/test_acc.npy',test_acc)
train_loss=np.array(train_loss)
np.save(path+'/train_loss.npy',train_loss)

200次迭代的结果如下：

cuda
/home/megstudio/_C10_Fixup_resnet
---  There is this folder!  ---
---  Create a new model...  ---
Epoch [1/200], Step [100/391] Loss: 2.2983
Epoch [1/200], Step [200/391] Loss: 1.8458
Epoch [1/200], Step [300/391] Loss: 1.6850
Accuracy of the model on the test images: 36.19 %
use 0.0min9.558553695678711s
Epoch [2/200], Step [100/391] Loss: 1.4505
Epoch [2/200], Step [200/391] Loss: 1.3100
Epoch [2/200], Step [300/391] Loss: 1.3405
Accuracy of the model on the test images: 48.39 %
use 0.0min9.359257936477661s
Epoch [3/200], Step [100/391] Loss: 1.2956
Epoch [3/200], Step [200/391] Loss: 1.3691
Epoch [3/200], Step [300/391] Loss: 1.2619
Accuracy of the model on the test images: 60.59 %
use 0.0min9.313580513000488s
Epoch [4/200], Step [100/391] Loss: 1.0492
Epoch [4/200], Step [200/391] Loss: 1.0419
Epoch [4/200], Step [300/391] Loss: 0.9458
Accuracy of the model on the test images: 65.9 %
use 0.0min9.629400253295898s
Epoch [5/200], Step [100/391] Loss: 0.9349
Epoch [5/200], Step [200/391] Loss: 0.9757
Epoch [5/200], Step [300/391] Loss: 0.8299
Accuracy of the model on the test images: 69.05 %
use 0.0min9.6714346408844s
Epoch [6/200], Step [100/391] Loss: 0.8565
Epoch [6/200], Step [200/391] Loss: 0.9004
Epoch [6/200], Step [300/391] Loss: 0.8459
Accuracy of the model on the test images: 70.2 %
use 0.0min9.550548553466797s
Epoch [7/200], Step [100/391] Loss: 0.6585
Epoch [7/200], Step [200/391] Loss: 0.9619
Epoch [7/200], Step [300/391] Loss: 0.7876
Accuracy of the model on the test images: 71.46 %
use 0.0min9.274351596832275s
Epoch [8/200], Step [100/391] Loss: 0.7694
Epoch [8/200], Step [200/391] Loss: 0.8057
Epoch [8/200], Step [300/391] Loss: 0.7867
Accuracy of the model on the test images: 71.77 %
use 0.0min9.717536687850952s
Epoch [9/200], Step [100/391] Loss: 0.7192
Epoch [9/200], Step [200/391] Loss: 0.7052
Epoch [9/200], Step [300/391] Loss: 0.6672
Accuracy of the model on the test images: 77.64 %
use 0.0min9.37049388885498s
Epoch [10/200], Step [100/391] Loss: 0.6820
Epoch [10/200], Step [200/391] Loss: 0.7275
Epoch [10/200], Step [300/391] Loss: 0.6165
Accuracy of the model on the test images: 76.81 %
use 0.0min9.321155309677124s
Epoch [11/200], Step [100/391] Loss: 0.7464
Epoch [11/200], Step [200/391] Loss: 0.6333
Epoch [11/200], Step [300/391] Loss: 0.5076
Accuracy of the model on the test images: 78.37 %
use 0.0min9.354505777359009s
Epoch [12/200], Step [100/391] Loss: 0.5930
Epoch [12/200], Step [200/391] Loss: 0.5188
Epoch [12/200], Step [300/391] Loss: 0.6838
Accuracy of the model on the test images: 79.38 %
use 0.0min9.66663646697998s
Epoch [13/200], Step [100/391] Loss: 0.5064
Epoch [13/200], Step [200/391] Loss: 0.5576
Epoch [13/200], Step [300/391] Loss: 0.6023
Accuracy of the model on the test images: 78.85 %
use 0.0min9.464256525039673s
Epoch [14/200], Step [100/391] Loss: 0.4986
Epoch [14/200], Step [200/391] Loss: 0.5994
Epoch [14/200], Step [300/391] Loss: 0.6045
Accuracy of the model on the test images: 79.54 %
use 0.0min9.234196424484253s
Epoch [15/200], Step [100/391] Loss: 0.4478
Epoch [15/200], Step [200/391] Loss: 0.5881
Epoch [15/200], Step [300/391] Loss: 0.6394
Accuracy of the model on the test images: 78.83 %
use 0.0min9.380087614059448s
Epoch [16/200], Step [100/391] Loss: 0.5672
Epoch [16/200], Step [200/391] Loss: 0.5794
Epoch [16/200], Step [300/391] Loss: 0.4188
Accuracy of the model on the test images: 79.93 %
use 0.0min9.55267596244812s
Epoch [17/200], Step [100/391] Loss: 0.6179
Epoch [17/200], Step [200/391] Loss: 0.4057
Epoch [17/200], Step [300/391] Loss: 0.5535
Accuracy of the model on the test images: 80.95 %
use 0.0min9.567511320114136s
Epoch [18/200], Step [100/391] Loss: 0.5181
Epoch [18/200], Step [200/391] Loss: 0.3983
Epoch [18/200], Step [300/391] Loss: 0.5233
Accuracy of the model on the test images: 79.61 %
use 0.0min9.399615287780762s
Epoch [19/200], Step [100/391] Loss: 0.6439
Epoch [19/200], Step [200/391] Loss: 0.4130
Epoch [19/200], Step [300/391] Loss: 0.3664
Accuracy of the model on the test images: 81.37 %
use 0.0min9.42436671257019s
Epoch [20/200], Step [100/391] Loss: 0.5308
Epoch [20/200], Step [200/391] Loss: 0.5670
Epoch [20/200], Step [300/391] Loss: 0.4917
Accuracy of the model on the test images: 78.91 %
use 0.0min9.535740375518799s
Epoch [21/200], Step [100/391] Loss: 0.6816
Epoch [21/200], Step [200/391] Loss: 0.5061
Epoch [21/200], Step [300/391] Loss: 0.6099
Accuracy of the model on the test images: 81.08 %
use 0.0min9.365344524383545s
Epoch [22/200], Step [100/391] Loss: 0.4436
Epoch [22/200], Step [200/391] Loss: 0.3794
Epoch [22/200], Step [300/391] Loss: 0.6551
Accuracy of the model on the test images: 82.04 %
use 0.0min9.271393060684204s
Epoch [23/200], Step [100/391] Loss: 0.4860
Epoch [23/200], Step [200/391] Loss: 0.4397
Epoch [23/200], Step [300/391] Loss: 0.5677
Accuracy of the model on the test images: 81.97 %
use 0.0min9.360072374343872s
Epoch [24/200], Step [100/391] Loss: 0.4033
Epoch [24/200], Step [200/391] Loss: 0.4754
Epoch [24/200], Step [300/391] Loss: 0.4589
Accuracy of the model on the test images: 81.41 %
use 0.0min9.276439189910889s
Epoch [25/200], Step [100/391] Loss: 0.4794
Epoch [25/200], Step [200/391] Loss: 0.5721
Epoch [25/200], Step [300/391] Loss: 0.5304
Accuracy of the model on the test images: 82.24 %
use 0.0min9.316136360168457s
Epoch [26/200], Step [100/391] Loss: 0.4591
Epoch [26/200], Step [200/391] Loss: 0.4771
Epoch [26/200], Step [300/391] Loss: 0.5548
Accuracy of the model on the test images: 82.87 %
use 0.0min9.299479722976685s
Epoch [27/200], Step [100/391] Loss: 0.4357
Epoch [27/200], Step [200/391] Loss: 0.5017
Epoch [27/200], Step [300/391] Loss: 0.5629
Accuracy of the model on the test images: 83.08 %
use 0.0min9.62153148651123s
Epoch [28/200], Step [100/391] Loss: 0.4526
Epoch [28/200], Step [200/391] Loss: 0.5035
Epoch [28/200], Step [300/391] Loss: 0.3142
Accuracy of the model on the test images: 82.46 %
use 0.0min9.602237224578857s
Epoch [29/200], Step [100/391] Loss: 0.3425
Epoch [29/200], Step [200/391] Loss: 0.4752
Epoch [29/200], Step [300/391] Loss: 0.4512
Accuracy of the model on the test images: 83.61 %
use 0.0min9.669076204299927s
Epoch [30/200], Step [100/391] Loss: 0.4429
Epoch [30/200], Step [200/391] Loss: 0.4209
Epoch [30/200], Step [300/391] Loss: 0.4763
Accuracy of the model on the test images: 83.57 %
use 0.0min9.29560375213623s
Epoch [31/200], Step [100/391] Loss: 0.5409
Epoch [31/200], Step [200/391] Loss: 0.3829
Epoch [31/200], Step [300/391] Loss: 0.6488
Accuracy of the model on the test images: 81.9 %
use 0.0min9.539057731628418s
Epoch [32/200], Step [100/391] Loss: 0.5364
Epoch [32/200], Step [200/391] Loss: 0.4393
Epoch [32/200], Step [300/391] Loss: 0.4632
Accuracy of the model on the test images: 82.38 %
use 0.0min9.281789541244507s
Epoch [33/200], Step [100/391] Loss: 0.4756
Epoch [33/200], Step [200/391] Loss: 0.4192
Epoch [33/200], Step [300/391] Loss: 0.4591
Accuracy of the model on the test images: 82.44 %
use 0.0min9.48759651184082s
Epoch [34/200], Step [100/391] Loss: 0.4581
Epoch [34/200], Step [200/391] Loss: 0.4677
Epoch [34/200], Step [300/391] Loss: 0.4400
Accuracy of the model on the test images: 79.7 %
use 0.0min9.844014644622803s
Epoch [35/200], Step [100/391] Loss: 0.4975
Epoch [35/200], Step [200/391] Loss: 0.4854
Epoch [35/200], Step [300/391] Loss: 0.4945
Accuracy of the model on the test images: 82.67 %
use 0.0min9.483444929122925s
Epoch [36/200], Step [100/391] Loss: 0.5306
Epoch [36/200], Step [200/391] Loss: 0.5698
Epoch [36/200], Step [300/391] Loss: 0.3759
Accuracy of the model on the test images: 82.85 %
use 0.0min9.318516254425049s
Epoch [37/200], Step [100/391] Loss: 0.5505
Epoch [37/200], Step [200/391] Loss: 0.5890
Epoch [37/200], Step [300/391] Loss: 0.4186
Accuracy of the model on the test images: 83.33 %
use 0.0min9.397469520568848s
Epoch [38/200], Step [100/391] Loss: 0.5184
Epoch [38/200], Step [200/391] Loss: 0.4628
Epoch [38/200], Step [300/391] Loss: 0.4720
Accuracy of the model on the test images: 83.6 %
use 0.0min9.332150220870972s
Epoch [39/200], Step [100/391] Loss: 0.5308
Epoch [39/200], Step [200/391] Loss: 0.5511
Epoch [39/200], Step [300/391] Loss: 0.4897
Accuracy of the model on the test images: 80.92 %
use 0.0min9.365912199020386s
Epoch [40/200], Step [100/391] Loss: 0.4245
Epoch [40/200], Step [200/391] Loss: 0.4502
Epoch [40/200], Step [300/391] Loss: 0.4129
Accuracy of the model on the test images: 81.82 %
use 0.0min9.359995365142822s
Epoch [41/200], Step [100/391] Loss: 0.4962
Epoch [41/200], Step [200/391] Loss: 0.4081
Epoch [41/200], Step [300/391] Loss: 0.4919
Accuracy of the model on the test images: 81.66 %
use 0.0min9.525192499160767s
Epoch [42/200], Step [100/391] Loss: 0.4365
Epoch [42/200], Step [200/391] Loss: 0.3061
Epoch [42/200], Step [300/391] Loss: 0.4191
Accuracy of the model on the test images: 83.57 %
use 0.0min9.527837991714478s
Epoch [43/200], Step [100/391] Loss: 0.5545
Epoch [43/200], Step [200/391] Loss: 0.5377
Epoch [43/200], Step [300/391] Loss: 0.5360
Accuracy of the model on the test images: 82.47 %
use 0.0min9.577733278274536s
Epoch [44/200], Step [100/391] Loss: 0.4107
Epoch [44/200], Step [200/391] Loss: 0.4287
Epoch [44/200], Step [300/391] Loss: 0.4797
Accuracy of the model on the test images: 83.19 %
use 0.0min9.359985113143921s
Epoch [45/200], Step [100/391] Loss: 0.4690
Epoch [45/200], Step [200/391] Loss: 0.4516
Epoch [45/200], Step [300/391] Loss: 0.4803
Accuracy of the model on the test images: 82.62 %
use 0.0min9.307728052139282s
Epoch [46/200], Step [100/391] Loss: 0.4665
Epoch [46/200], Step [200/391] Loss: 0.4064
Epoch [46/200], Step [300/391] Loss: 0.5063
Accuracy of the model on the test images: 82.64 %
use 0.0min9.411344289779663s
Epoch [47/200], Step [100/391] Loss: 0.3875
Epoch [47/200], Step [200/391] Loss: 0.5315
Epoch [47/200], Step [300/391] Loss: 0.3919
Accuracy of the model on the test images: 83.08 %
use 0.0min9.434369325637817s
Epoch [48/200], Step [100/391] Loss: 0.4308
Epoch [48/200], Step [200/391] Loss: 0.4459
Epoch [48/200], Step [300/391] Loss: 0.3898
Accuracy of the model on the test images: 81.31 %
use 0.0min9.367540121078491s
Epoch [49/200], Step [100/391] Loss: 0.3662
Epoch [49/200], Step [200/391] Loss: 0.5421
Epoch [49/200], Step [300/391] Loss: 0.4911
Accuracy of the model on the test images: 84.19 %
use 0.0min9.508936643600464s
Epoch [50/200], Step [100/391] Loss: 0.3433
Epoch [50/200], Step [200/391] Loss: 0.4138
Epoch [50/200], Step [300/391] Loss: 0.3959
Accuracy of the model on the test images: 83.98 %
use 0.0min9.390262603759766s
Epoch [51/200], Step [100/391] Loss: 0.4110
Epoch [51/200], Step [200/391] Loss: 0.5862
Epoch [51/200], Step [300/391] Loss: 0.5232
Accuracy of the model on the test images: 83.84 %
use 0.0min9.442183256149292s
Epoch [52/200], Step [100/391] Loss: 0.5054
Epoch [52/200], Step [200/391] Loss: 0.5269
Epoch [52/200], Step [300/391] Loss: 0.4547
Accuracy of the model on the test images: 82.66 %
use 0.0min9.26649260520935s
Epoch [53/200], Step [100/391] Loss: 0.4397
Epoch [53/200], Step [200/391] Loss: 0.3480
Epoch [53/200], Step [300/391] Loss: 0.5525
Accuracy of the model on the test images: 84.36 %
use 0.0min9.58961534500122s
Epoch [54/200], Step [100/391] Loss: 0.2681
Epoch [54/200], Step [200/391] Loss: 0.4127
Epoch [54/200], Step [300/391] Loss: 0.5100
Accuracy of the model on the test images: 83.57 %
use 0.0min9.334172010421753s
Epoch [55/200], Step [100/391] Loss: 0.4540
Epoch [55/200], Step [200/391] Loss: 0.3190
Epoch [55/200], Step [300/391] Loss: 0.3831
Accuracy of the model on the test images: 84.38 %
use 0.0min9.314176082611084s
Epoch [56/200], Step [100/391] Loss: 0.4402
Epoch [56/200], Step [200/391] Loss: 0.5519
Epoch [56/200], Step [300/391] Loss: 0.5221
Accuracy of the model on the test images: 84.28 %
use 0.0min9.577316045761108s
Epoch [57/200], Step [100/391] Loss: 0.4109
Epoch [57/200], Step [200/391] Loss: 0.4810
Epoch [57/200], Step [300/391] Loss: 0.4875
Accuracy of the model on the test images: 82.6 %
use 0.0min9.43066692352295s
Epoch [58/200], Step [100/391] Loss: 0.3342
Epoch [58/200], Step [200/391] Loss: 0.3503
Epoch [58/200], Step [300/391] Loss: 0.4294
Accuracy of the model on the test images: 84.36 %
use 0.0min9.546979665756226s
Epoch [59/200], Step [100/391] Loss: 0.4277
Epoch [59/200], Step [200/391] Loss: 0.4011
Epoch [59/200], Step [300/391] Loss: 0.4361
Accuracy of the model on the test images: 83.79 %
use 0.0min9.404965162277222s
Epoch [60/200], Step [100/391] Loss: 0.3305
Epoch [60/200], Step [200/391] Loss: 0.5688
Epoch [60/200], Step [300/391] Loss: 0.5261
Accuracy of the model on the test images: 84.48 %
use 0.0min9.559430360794067s
Epoch [61/200], Step [100/391] Loss: 0.4395
Epoch [61/200], Step [200/391] Loss: 0.3885
Epoch [61/200], Step [300/391] Loss: 0.5140
Accuracy of the model on the test images: 83.49 %
use 0.0min9.369972467422485s
Epoch [62/200], Step [100/391] Loss: 0.4011
Epoch [62/200], Step [200/391] Loss: 0.3722
Epoch [62/200], Step [300/391] Loss: 0.5032
Accuracy of the model on the test images: 84.01 %
use 0.0min9.315366983413696s
Epoch [63/200], Step [100/391] Loss: 0.4591
Epoch [63/200], Step [200/391] Loss: 0.4608
Epoch [63/200], Step [300/391] Loss: 0.3881
Accuracy of the model on the test images: 84.86 %
use 0.0min9.548750400543213s
Epoch [64/200], Step [100/391] Loss: 0.4841
Epoch [64/200], Step [200/391] Loss: 0.3501
Epoch [64/200], Step [300/391] Loss: 0.4830
Accuracy of the model on the test images: 84.16 %
use 0.0min9.573681831359863s
Epoch [65/200], Step [100/391] Loss: 0.5717
Epoch [65/200], Step [200/391] Loss: 0.3605
Epoch [65/200], Step [300/391] Loss: 0.3855
Accuracy of the model on the test images: 83.49 %
use 0.0min9.306811809539795s
Epoch [66/200], Step [100/391] Loss: 0.4324
Epoch [66/200], Step [200/391] Loss: 0.3139
Epoch [66/200], Step [300/391] Loss: 0.5673
Accuracy of the model on the test images: 83.57 %
use 0.0min9.49308466911316s
Epoch [67/200], Step [100/391] Loss: 0.3916
Epoch [67/200], Step [200/391] Loss: 0.3329
Epoch [67/200], Step [300/391] Loss: 0.5123
Accuracy of the model on the test images: 84.18 %
use 0.0min9.484142780303955s
Epoch [68/200], Step [100/391] Loss: 0.4350
Epoch [68/200], Step [200/391] Loss: 0.4689
Epoch [68/200], Step [300/391] Loss: 0.3213
Accuracy of the model on the test images: 82.39 %
use 0.0min9.45761513710022s
Epoch [69/200], Step [100/391] Loss: 0.5564
Epoch [69/200], Step [200/391] Loss: 0.4594
Epoch [69/200], Step [300/391] Loss: 0.3615
Accuracy of the model on the test images: 84.55 %
use 0.0min9.39707636833191s
Epoch [70/200], Step [100/391] Loss: 0.3714
Epoch [70/200], Step [200/391] Loss: 0.4805
Epoch [70/200], Step [300/391] Loss: 0.5235
Accuracy of the model on the test images: 84.34 %
use 0.0min9.40149712562561s
Epoch [71/200], Step [100/391] Loss: 0.5379
Epoch [71/200], Step [200/391] Loss: 0.4112
Epoch [71/200], Step [300/391] Loss: 0.4575
Accuracy of the model on the test images: 83.94 %
use 0.0min9.816120624542236s
Epoch [72/200], Step [100/391] Loss: 0.5363
Epoch [72/200], Step [200/391] Loss: 0.4670
Epoch [72/200], Step [300/391] Loss: 0.3321
Accuracy of the model on the test images: 84.24 %
use 0.0min9.441385984420776s
Epoch [73/200], Step [100/391] Loss: 0.6014
Epoch [73/200], Step [200/391] Loss: 0.3428
Epoch [73/200], Step [300/391] Loss: 0.4581
Accuracy of the model on the test images: 84.49 %
use 0.0min9.380602359771729s
Epoch [74/200], Step [100/391] Loss: 0.2939
Epoch [74/200], Step [200/391] Loss: 0.3973
Epoch [74/200], Step [300/391] Loss: 0.4490
Accuracy of the model on the test images: 84.75 %
use 0.0min9.382431745529175s
Epoch [75/200], Step [100/391] Loss: 0.4135
Epoch [75/200], Step [200/391] Loss: 0.4183
Epoch [75/200], Step [300/391] Loss: 0.4768
Accuracy of the model on the test images: 82.78 %
use 0.0min9.409318923950195s
Epoch [76/200], Step [100/391] Loss: 0.3885
Epoch [76/200], Step [200/391] Loss: 0.3819
Epoch [76/200], Step [300/391] Loss: 0.5817
Accuracy of the model on the test images: 84.37 %
use 0.0min9.35073733329773s
Epoch [77/200], Step [100/391] Loss: 0.5361
Epoch [77/200], Step [200/391] Loss: 0.4904
Epoch [77/200], Step [300/391] Loss: 0.3602
Accuracy of the model on the test images: 85.2 %
use 0.0min9.365845918655396s
Epoch [78/200], Step [100/391] Loss: 0.4182
Epoch [78/200], Step [200/391] Loss: 0.4124
Epoch [78/200], Step [300/391] Loss: 0.4921
Accuracy of the model on the test images: 84.34 %
use 0.0min9.939101934432983s
Epoch [79/200], Step [100/391] Loss: 0.3274
Epoch [79/200], Step [200/391] Loss: 0.5387
Epoch [79/200], Step [300/391] Loss: 0.3468
Accuracy of the model on the test images: 84.44 %
use 0.0min9.403112888336182s
Epoch [80/200], Step [100/391] Loss: 0.3899
Epoch [80/200], Step [200/391] Loss: 0.4155
Epoch [80/200], Step [300/391] Loss: 0.2920
Accuracy of the model on the test images: 84.35 %
use 0.0min9.479256868362427s
Epoch [81/200], Step [100/391] Loss: 0.3786
Epoch [81/200], Step [200/391] Loss: 0.4750
Epoch [81/200], Step [300/391] Loss: 0.4865
Accuracy of the model on the test images: 84.62 %
use 0.0min9.400079011917114s
Epoch [82/200], Step [100/391] Loss: 0.4530
Epoch [82/200], Step [200/391] Loss: 0.4651
Epoch [82/200], Step [300/391] Loss: 0.2479
Accuracy of the model on the test images: 82.71 %
use 0.0min9.616268157958984s
Epoch [83/200], Step [100/391] Loss: 0.4265
Epoch [83/200], Step [200/391] Loss: 0.3879
Epoch [83/200], Step [300/391] Loss: 0.5497
Accuracy of the model on the test images: 85.13 %
use 0.0min9.36326789855957s
Epoch [84/200], Step [100/391] Loss: 0.6209
Epoch [84/200], Step [200/391] Loss: 0.3982
Epoch [84/200], Step [300/391] Loss: 0.3876
Accuracy of the model on the test images: 83.28 %
use 0.0min9.573141098022461s
Epoch [85/200], Step [100/391] Loss: 0.4384
Epoch [85/200], Step [200/391] Loss: 0.3740
Epoch [85/200], Step [300/391] Loss: 0.5753
Accuracy of the model on the test images: 84.71 %
use 0.0min9.812830924987793s
Epoch [86/200], Step [100/391] Loss: 0.4309
Epoch [86/200], Step [200/391] Loss: 0.3098
Epoch [86/200], Step [300/391] Loss: 0.4326
Accuracy of the model on the test images: 83.57 %
use 0.0min9.380079746246338s
Epoch [87/200], Step [100/391] Loss: 0.5606
Epoch [87/200], Step [200/391] Loss: 0.5205
Epoch [87/200], Step [300/391] Loss: 0.4379
Accuracy of the model on the test images: 83.51 %
use 0.0min9.353623390197754s
Epoch [88/200], Step [100/391] Loss: 0.4372
Epoch [88/200], Step [200/391] Loss: 0.5361
Epoch [88/200], Step [300/391] Loss: 0.5414
Accuracy of the model on the test images: 83.62 %
use 0.0min9.28493046760559s
Epoch [89/200], Step [100/391] Loss: 0.4471
Epoch [89/200], Step [200/391] Loss: 0.4187
Epoch [89/200], Step [300/391] Loss: 0.4998
Accuracy of the model on the test images: 82.83 %
use 0.0min9.424630641937256s
Epoch [90/200], Step [100/391] Loss: 0.4585
Epoch [90/200], Step [200/391] Loss: 0.3312
Epoch [90/200], Step [300/391] Loss: 0.5149
Accuracy of the model on the test images: 82.93 %
use 0.0min9.5675790309906s
Epoch [91/200], Step [100/391] Loss: 0.3778
Epoch [91/200], Step [200/391] Loss: 0.3842
Epoch [91/200], Step [300/391] Loss: 0.5968
Accuracy of the model on the test images: 83.06 %
use 0.0min9.773025274276733s
Epoch [92/200], Step [100/391] Loss: 0.4986
Epoch [92/200], Step [200/391] Loss: 0.3446
Epoch [92/200], Step [300/391] Loss: 0.4200
Accuracy of the model on the test images: 84.89 %
use 0.0min9.527753114700317s
Epoch [93/200], Step [100/391] Loss: 0.3094
Epoch [93/200], Step [200/391] Loss: 0.3540
Epoch [93/200], Step [300/391] Loss: 0.3939
Accuracy of the model on the test images: 85.52 %
use 0.0min9.599644660949707s
Epoch [94/200], Step [100/391] Loss: 0.3335
Epoch [94/200], Step [200/391] Loss: 0.3995
Epoch [94/200], Step [300/391] Loss: 0.4208
Accuracy of the model on the test images: 83.76 %
use 0.0min9.283499956130981s
Epoch [95/200], Step [100/391] Loss: 0.2264
Epoch [95/200], Step [200/391] Loss: 0.4540
Epoch [95/200], Step [300/391] Loss: 0.4403
Accuracy of the model on the test images: 85.06 %
use 0.0min9.470134496688843s
Epoch [96/200], Step [100/391] Loss: 0.3941
Epoch [96/200], Step [200/391] Loss: 0.3983
Epoch [96/200], Step [300/391] Loss: 0.5474
Accuracy of the model on the test images: 84.1 %
use 0.0min9.545747518539429s
Epoch [97/200], Step [100/391] Loss: 0.4523
Epoch [97/200], Step [200/391] Loss: 0.5153
Epoch [97/200], Step [300/391] Loss: 0.3140
Accuracy of the model on the test images: 84.65 %
use 0.0min9.316319227218628s
Epoch [98/200], Step [100/391] Loss: 0.4368
Epoch [98/200], Step [200/391] Loss: 0.3617
Epoch [98/200], Step [300/391] Loss: 0.3359
Accuracy of the model on the test images: 83.95 %
use 0.0min9.305066108703613s
Epoch [99/200], Step [100/391] Loss: 0.6148
Epoch [99/200], Step [200/391] Loss: 0.5048
Epoch [99/200], Step [300/391] Loss: 0.5112
Accuracy of the model on the test images: 83.91 %
use 0.0min9.295705318450928s
Epoch [100/200], Step [100/391] Loss: 0.4258
Epoch [100/200], Step [200/391] Loss: 0.2739
Epoch [100/200], Step [300/391] Loss: 0.2184
Accuracy of the model on the test images: 89.67 %
use 0.0min9.649784326553345s
Epoch [101/200], Step [100/391] Loss: 0.1777
Epoch [101/200], Step [200/391] Loss: 0.2015
Epoch [101/200], Step [300/391] Loss: 0.1458
Accuracy of the model on the test images: 90.01 %
use 0.0min9.405852794647217s
Epoch [102/200], Step [100/391] Loss: 0.2960
Epoch [102/200], Step [200/391] Loss: 0.1928
Epoch [102/200], Step [300/391] Loss: 0.1295
Accuracy of the model on the test images: 90.49 %
use 0.0min9.48672866821289s
Epoch [103/200], Step [100/391] Loss: 0.2136
Epoch [103/200], Step [200/391] Loss: 0.2179
Epoch [103/200], Step [300/391] Loss: 0.1613
Accuracy of the model on the test images: 90.44 %
use 0.0min9.369870901107788s
Epoch [104/200], Step [100/391] Loss: 0.1496
Epoch [104/200], Step [200/391] Loss: 0.0888
Epoch [104/200], Step [300/391] Loss: 0.1923
Accuracy of the model on the test images: 90.56 %
use 0.0min9.41650128364563s
Epoch [105/200], Step [100/391] Loss: 0.1235
Epoch [105/200], Step [200/391] Loss: 0.1581
Epoch [105/200], Step [300/391] Loss: 0.2020
Accuracy of the model on the test images: 90.87 %
use 0.0min9.425810098648071s
Epoch [106/200], Step [100/391] Loss: 0.0974
Epoch [106/200], Step [200/391] Loss: 0.2375
Epoch [106/200], Step [300/391] Loss: 0.1368
Accuracy of the model on the test images: 90.73 %
use 0.0min9.438855171203613s
Epoch [107/200], Step [100/391] Loss: 0.1596
Epoch [107/200], Step [200/391] Loss: 0.1657
Epoch [107/200], Step [300/391] Loss: 0.0694
Accuracy of the model on the test images: 90.88 %
use 0.0min9.617606401443481s
Epoch [108/200], Step [100/391] Loss: 0.0887
Epoch [108/200], Step [200/391] Loss: 0.1985
Epoch [108/200], Step [300/391] Loss: 0.0848
Accuracy of the model on the test images: 90.9 %
use 0.0min9.355462074279785s
Epoch [109/200], Step [100/391] Loss: 0.0991
Epoch [109/200], Step [200/391] Loss: 0.0786
Epoch [109/200], Step [300/391] Loss: 0.0753
Accuracy of the model on the test images: 90.75 %
use 0.0min9.562483072280884s
Epoch [110/200], Step [100/391] Loss: 0.0743
Epoch [110/200], Step [200/391] Loss: 0.1927
Epoch [110/200], Step [300/391] Loss: 0.0883
Accuracy of the model on the test images: 90.94 %
use 0.0min9.486653804779053s
Epoch [111/200], Step [100/391] Loss: 0.1103
Epoch [111/200], Step [200/391] Loss: 0.1331
Epoch [111/200], Step [300/391] Loss: 0.0378
Accuracy of the model on the test images: 90.94 %
use 0.0min9.580935955047607s
Epoch [112/200], Step [100/391] Loss: 0.0681
Epoch [112/200], Step [200/391] Loss: 0.1717
Epoch [112/200], Step [300/391] Loss: 0.0652
Accuracy of the model on the test images: 91.02 %
use 0.0min9.371098041534424s
Epoch [113/200], Step [100/391] Loss: 0.1344
Epoch [113/200], Step [200/391] Loss: 0.0706
Epoch [113/200], Step [300/391] Loss: 0.0870
Accuracy of the model on the test images: 90.93 %
use 0.0min9.35228943824768s
Epoch [114/200], Step [100/391] Loss: 0.1175
Epoch [114/200], Step [200/391] Loss: 0.0625
Epoch [114/200], Step [300/391] Loss: 0.0716
Accuracy of the model on the test images: 90.96 %
use 0.0min9.561415433883667s
Epoch [115/200], Step [100/391] Loss: 0.1224
Epoch [115/200], Step [200/391] Loss: 0.1254
Epoch [115/200], Step [300/391] Loss: 0.0313
Accuracy of the model on the test images: 91.05 %
use 0.0min9.569977760314941s
Epoch [116/200], Step [100/391] Loss: 0.1271
Epoch [116/200], Step [200/391] Loss: 0.0848
Epoch [116/200], Step [300/391] Loss: 0.0753
Accuracy of the model on the test images: 90.76 %
use 0.0min9.487900257110596s
Epoch [117/200], Step [100/391] Loss: 0.0537
Epoch [117/200], Step [200/391] Loss: 0.1041
Epoch [117/200], Step [300/391] Loss: 0.1030
Accuracy of the model on the test images: 90.69 %
use 0.0min9.460529804229736s
Epoch [118/200], Step [100/391] Loss: 0.1120
Epoch [118/200], Step [200/391] Loss: 0.1395
Epoch [118/200], Step [300/391] Loss: 0.0407
Accuracy of the model on the test images: 90.75 %
use 0.0min9.484214305877686s
Epoch [119/200], Step [100/391] Loss: 0.1268
Epoch [119/200], Step [200/391] Loss: 0.0956
Epoch [119/200], Step [300/391] Loss: 0.0616
Accuracy of the model on the test images: 90.3 %
use 0.0min9.527601718902588s
Epoch [120/200], Step [100/391] Loss: 0.1024
Epoch [120/200], Step [200/391] Loss: 0.0360
Epoch [120/200], Step [300/391] Loss: 0.0755
Accuracy of the model on the test images: 91.22 %
use 0.0min9.299079656600952s
Epoch [121/200], Step [100/391] Loss: 0.0659
Epoch [121/200], Step [200/391] Loss: 0.0670
Epoch [121/200], Step [300/391] Loss: 0.0510
Accuracy of the model on the test images: 90.98 %
use 0.0min9.58971881866455s
Epoch [122/200], Step [100/391] Loss: 0.0236
Epoch [122/200], Step [200/391] Loss: 0.0559
Epoch [122/200], Step [300/391] Loss: 0.0943
Accuracy of the model on the test images: 90.93 %
use 0.0min9.642295122146606s
Epoch [123/200], Step [100/391] Loss: 0.0856
Epoch [123/200], Step [200/391] Loss: 0.1188
Epoch [123/200], Step [300/391] Loss: 0.0428
Accuracy of the model on the test images: 90.77 %
use 0.0min9.619181871414185s
Epoch [124/200], Step [100/391] Loss: 0.1194
Epoch [124/200], Step [200/391] Loss: 0.0595
Epoch [124/200], Step [300/391] Loss: 0.0497
Accuracy of the model on the test images: 90.65 %
use 0.0min9.483120679855347s
Epoch [125/200], Step [100/391] Loss: 0.0647
Epoch [125/200], Step [200/391] Loss: 0.0639
Epoch [125/200], Step [300/391] Loss: 0.0387
Accuracy of the model on the test images: 90.54 %
use 0.0min9.407922983169556s
Epoch [126/200], Step [100/391] Loss: 0.0452
Epoch [126/200], Step [200/391] Loss: 0.0783
Epoch [126/200], Step [300/391] Loss: 0.0932
Accuracy of the model on the test images: 90.65 %
use 0.0min9.325565099716187s
Epoch [127/200], Step [100/391] Loss: 0.0463
Epoch [127/200], Step [200/391] Loss: 0.0480
Epoch [127/200], Step [300/391] Loss: 0.0901
Accuracy of the model on the test images: 90.79 %
use 0.0min9.477588653564453s
Epoch [128/200], Step [100/391] Loss: 0.0690
Epoch [128/200], Step [200/391] Loss: 0.0372
Epoch [128/200], Step [300/391] Loss: 0.1325
Accuracy of the model on the test images: 90.49 %
use 0.0min9.388424396514893s
Epoch [129/200], Step [100/391] Loss: 0.0436
Epoch [129/200], Step [200/391] Loss: 0.1335
Epoch [129/200], Step [300/391] Loss: 0.0868
Accuracy of the model on the test images: 90.52 %
use 0.0min9.61071252822876s
Epoch [130/200], Step [100/391] Loss: 0.0403
Epoch [130/200], Step [200/391] Loss: 0.1448
Epoch [130/200], Step [300/391] Loss: 0.1000
Accuracy of the model on the test images: 89.94 %
use 0.0min9.41750168800354s
Epoch [131/200], Step [100/391] Loss: 0.0772
Epoch [131/200], Step [200/391] Loss: 0.0466
Epoch [131/200], Step [300/391] Loss: 0.0327
Accuracy of the model on the test images: 90.38 %
use 0.0min9.277085304260254s
Epoch [132/200], Step [100/391] Loss: 0.0382
Epoch [132/200], Step [200/391] Loss: 0.0377
Epoch [132/200], Step [300/391] Loss: 0.1461
Accuracy of the model on the test images: 90.61 %
use 0.0min9.435935258865356s
Epoch [133/200], Step [100/391] Loss: 0.0933
Epoch [133/200], Step [200/391] Loss: 0.0637
Epoch [133/200], Step [300/391] Loss: 0.0430
Accuracy of the model on the test images: 90.51 %
use 0.0min9.535774230957031s
Epoch [134/200], Step [100/391] Loss: 0.0857
Epoch [134/200], Step [200/391] Loss: 0.0480
Epoch [134/200], Step [300/391] Loss: 0.0571
Accuracy of the model on the test images: 90.91 %
use 0.0min9.312682867050171s
Epoch [135/200], Step [100/391] Loss: 0.0793
Epoch [135/200], Step [200/391] Loss: 0.0878
Epoch [135/200], Step [300/391] Loss: 0.0669
Accuracy of the model on the test images: 90.12 %
use 0.0min9.361658096313477s
Epoch [136/200], Step [100/391] Loss: 0.0372
Epoch [136/200], Step [200/391] Loss: 0.0634
Epoch [136/200], Step [300/391] Loss: 0.0332
Accuracy of the model on the test images: 90.04 %
use 0.0min9.717036247253418s
Epoch [137/200], Step [100/391] Loss: 0.0508
Epoch [137/200], Step [200/391] Loss: 0.0714
Epoch [137/200], Step [300/391] Loss: 0.1258
Accuracy of the model on the test images: 90.46 %
use 0.0min9.424921035766602s
Epoch [138/200], Step [100/391] Loss: 0.1119
Epoch [138/200], Step [200/391] Loss: 0.0357
Epoch [138/200], Step [300/391] Loss: 0.0616
Accuracy of the model on the test images: 90.79 %
use 0.0min9.382011890411377s
Epoch [139/200], Step [100/391] Loss: 0.1467
Epoch [139/200], Step [200/391] Loss: 0.0231
Epoch [139/200], Step [300/391] Loss: 0.0705
Accuracy of the model on the test images: 90.62 %
use 0.0min9.59222149848938s
Epoch [140/200], Step [100/391] Loss: 0.0601
Epoch [140/200], Step [200/391] Loss: 0.0528
Epoch [140/200], Step [300/391] Loss: 0.1169
Accuracy of the model on the test images: 90.61 %
use 0.0min9.37021255493164s
Epoch [141/200], Step [100/391] Loss: 0.0626
Epoch [141/200], Step [200/391] Loss: 0.1285
Epoch [141/200], Step [300/391] Loss: 0.0525
Accuracy of the model on the test images: 90.9 %
use 0.0min9.31493592262268s
Epoch [142/200], Step [100/391] Loss: 0.0422
Epoch [142/200], Step [200/391] Loss: 0.0349
Epoch [142/200], Step [300/391] Loss: 0.0821
Accuracy of the model on the test images: 90.98 %
use 0.0min9.590577602386475s
Epoch [143/200], Step [100/391] Loss: 0.0407
Epoch [143/200], Step [200/391] Loss: 0.0981
Epoch [143/200], Step [300/391] Loss: 0.1042
Accuracy of the model on the test images: 90.88 %
use 0.0min9.73942494392395s
Epoch [144/200], Step [100/391] Loss: 0.0344
Epoch [144/200], Step [200/391] Loss: 0.0643
Epoch [144/200], Step [300/391] Loss: 0.0817
Accuracy of the model on the test images: 90.53 %
use 0.0min9.600599527359009s
Epoch [145/200], Step [100/391] Loss: 0.1076
Epoch [145/200], Step [200/391] Loss: 0.0638
Epoch [145/200], Step [300/391] Loss: 0.0750
Accuracy of the model on the test images: 90.62 %
use 0.0min9.641296863555908s
Epoch [146/200], Step [100/391] Loss: 0.0762
Epoch [146/200], Step [200/391] Loss: 0.0617
Epoch [146/200], Step [300/391] Loss: 0.1381
Accuracy of the model on the test images: 91.25 %
use 0.0min9.690308332443237s
Epoch [147/200], Step [100/391] Loss: 0.0355
Epoch [147/200], Step [200/391] Loss: 0.1418
Epoch [147/200], Step [300/391] Loss: 0.0825
Accuracy of the model on the test images: 90.33 %
use 0.0min9.586420059204102s
Epoch [148/200], Step [100/391] Loss: 0.1283
Epoch [148/200], Step [200/391] Loss: 0.0497
Epoch [148/200], Step [300/391] Loss: 0.1110
Accuracy of the model on the test images: 90.36 %
use 0.0min9.302953243255615s
Epoch [149/200], Step [100/391] Loss: 0.0635
Epoch [149/200], Step [200/391] Loss: 0.0681
Epoch [149/200], Step [300/391] Loss: 0.0403
Accuracy of the model on the test images: 90.64 %
use 0.0min9.380799531936646s
Epoch [150/200], Step [100/391] Loss: 0.0998
Epoch [150/200], Step [200/391] Loss: 0.0508
Epoch [150/200], Step [300/391] Loss: 0.0221
Accuracy of the model on the test images: 91.84 %
use 0.0min9.811650276184082s
Epoch [151/200], Step [100/391] Loss: 0.0184
Epoch [151/200], Step [200/391] Loss: 0.0229
Epoch [151/200], Step [300/391] Loss: 0.0444
Accuracy of the model on the test images: 92.11 %
use 0.0min9.56403136253357s
Epoch [152/200], Step [100/391] Loss: 0.0159
Epoch [152/200], Step [200/391] Loss: 0.0139
Epoch [152/200], Step [300/391] Loss: 0.0077
Accuracy of the model on the test images: 92.05 %
use 0.0min9.317204236984253s
Epoch [153/200], Step [100/391] Loss: 0.0134
Epoch [153/200], Step [200/391] Loss: 0.0380
Epoch [153/200], Step [300/391] Loss: 0.0265
Accuracy of the model on the test images: 91.92 %
use 0.0min9.460333347320557s
Epoch [154/200], Step [100/391] Loss: 0.0365
Epoch [154/200], Step [200/391] Loss: 0.1153
Epoch [154/200], Step [300/391] Loss: 0.0244
Accuracy of the model on the test images: 92.02 %
use 0.0min9.36781930923462s
Epoch [155/200], Step [100/391] Loss: 0.0443
Epoch [155/200], Step [200/391] Loss: 0.0077
Epoch [155/200], Step [300/391] Loss: 0.0115
Accuracy of the model on the test images: 92.0 %
use 0.0min9.26927399635315s
Epoch [156/200], Step [100/391] Loss: 0.0566
Epoch [156/200], Step [200/391] Loss: 0.0302
Epoch [156/200], Step [300/391] Loss: 0.0031
Accuracy of the model on the test images: 92.08 %
use 0.0min9.242826223373413s
Epoch [157/200], Step [100/391] Loss: 0.0078
Epoch [157/200], Step [200/391] Loss: 0.0594
Epoch [157/200], Step [300/391] Loss: 0.0061
Accuracy of the model on the test images: 92.14 %
use 0.0min9.523272037506104s
Epoch [158/200], Step [100/391] Loss: 0.0222
Epoch [158/200], Step [200/391] Loss: 0.0169
Epoch [158/200], Step [300/391] Loss: 0.0131
Accuracy of the model on the test images: 91.94 %
use 0.0min9.707300424575806s
Epoch [159/200], Step [100/391] Loss: 0.0190
Epoch [159/200], Step [200/391] Loss: 0.0096
Epoch [159/200], Step [300/391] Loss: 0.0093
Accuracy of the model on the test images: 92.21 %
use 0.0min9.292672157287598s
Epoch [160/200], Step [100/391] Loss: 0.0041
Epoch [160/200], Step [200/391] Loss: 0.0143
Epoch [160/200], Step [300/391] Loss: 0.0051
Accuracy of the model on the test images: 92.0 %
use 0.0min9.329062223434448s
Epoch [161/200], Step [100/391] Loss: 0.0128
Epoch [161/200], Step [200/391] Loss: 0.0114
Epoch [161/200], Step [300/391] Loss: 0.0016
Accuracy of the model on the test images: 91.96 %
use 0.0min9.321907043457031s
Epoch [162/200], Step [100/391] Loss: 0.0099
Epoch [162/200], Step [200/391] Loss: 0.0122
Epoch [162/200], Step [300/391] Loss: 0.0049
Accuracy of the model on the test images: 92.03 %
use 0.0min9.442137002944946s
Epoch [163/200], Step [100/391] Loss: 0.0037
Epoch [163/200], Step [200/391] Loss: 0.0272
Epoch [163/200], Step [300/391] Loss: 0.0059
Accuracy of the model on the test images: 92.09 %
use 0.0min9.420178651809692s
Epoch [164/200], Step [100/391] Loss: 0.0137
Epoch [164/200], Step [200/391] Loss: 0.0058
Epoch [164/200], Step [300/391] Loss: 0.0038
Accuracy of the model on the test images: 92.16 %
use 0.0min9.320745944976807s
Epoch [165/200], Step [100/391] Loss: 0.0084
Epoch [165/200], Step [200/391] Loss: 0.0157
Epoch [165/200], Step [300/391] Loss: 0.0078
Accuracy of the model on the test images: 92.3 %
use 0.0min9.725970029830933s
Epoch [166/200], Step [100/391] Loss: 0.0078
Epoch [166/200], Step [200/391] Loss: 0.0126
Epoch [166/200], Step [300/391] Loss: 0.0036
Accuracy of the model on the test images: 92.14 %
use 0.0min9.462528705596924s
Epoch [167/200], Step [100/391] Loss: 0.0050
Epoch [167/200], Step [200/391] Loss: 0.0038
Epoch [167/200], Step [300/391] Loss: 0.0127
Accuracy of the model on the test images: 92.12 %
use 0.0min9.33044958114624s
Epoch [168/200], Step [100/391] Loss: 0.0016
Epoch [168/200], Step [200/391] Loss: 0.0118
Epoch [168/200], Step [300/391] Loss: 0.0053
Accuracy of the model on the test images: 92.22 %
use 0.0min9.298505067825317s
Epoch [169/200], Step [100/391] Loss: 0.0016
Epoch [169/200], Step [200/391] Loss: 0.0019
Epoch [169/200], Step [300/391] Loss: 0.0031
Accuracy of the model on the test images: 92.19 %
use 0.0min9.409983396530151s
Epoch [170/200], Step [100/391] Loss: 0.0057
Epoch [170/200], Step [200/391] Loss: 0.0239
Epoch [170/200], Step [300/391] Loss: 0.0074
Accuracy of the model on the test images: 92.25 %
use 0.0min9.341278791427612s
Epoch [171/200], Step [100/391] Loss: 0.0045
Epoch [171/200], Step [200/391] Loss: 0.0039
Epoch [171/200], Step [300/391] Loss: 0.0047
Accuracy of the model on the test images: 92.19 %
use 0.0min9.469080924987793s
Epoch [172/200], Step [100/391] Loss: 0.0085
Epoch [172/200], Step [200/391] Loss: 0.0017
Epoch [172/200], Step [300/391] Loss: 0.0023
Accuracy of the model on the test images: 92.06 %
use 0.0min9.527499437332153s
Epoch [173/200], Step [100/391] Loss: 0.0072
Epoch [173/200], Step [200/391] Loss: 0.0200
Epoch [173/200], Step [300/391] Loss: 0.0051
Accuracy of the model on the test images: 92.24 %
use 0.0min9.564720392227173s
Epoch [174/200], Step [100/391] Loss: 0.0029
Epoch [174/200], Step [200/391] Loss: 0.0102
Epoch [174/200], Step [300/391] Loss: 0.0045
Accuracy of the model on the test images: 92.18 %
use 0.0min9.283840894699097s
Epoch [175/200], Step [100/391] Loss: 0.0040
Epoch [175/200], Step [200/391] Loss: 0.0015
Epoch [175/200], Step [300/391] Loss: 0.0145
Accuracy of the model on the test images: 92.25 %
use 0.0min9.369336605072021s
Epoch [176/200], Step [100/391] Loss: 0.0066
Epoch [176/200], Step [200/391] Loss: 0.0045
Epoch [176/200], Step [300/391] Loss: 0.0747
Accuracy of the model on the test images: 92.29 %
use 0.0min9.734862565994263s
Epoch [177/200], Step [100/391] Loss: 0.0069
Epoch [177/200], Step [200/391] Loss: 0.0019
Epoch [177/200], Step [300/391] Loss: 0.0071
Accuracy of the model on the test images: 92.25 %
use 0.0min9.56549334526062s
Epoch [178/200], Step [100/391] Loss: 0.0177
Epoch [178/200], Step [200/391] Loss: 0.0053
Epoch [178/200], Step [300/391] Loss: 0.0376
Accuracy of the model on the test images: 92.27 %
use 0.0min9.20121431350708s
Epoch [179/200], Step [100/391] Loss: 0.0039
Epoch [179/200], Step [200/391] Loss: 0.0022
Epoch [179/200], Step [300/391] Loss: 0.0034
Accuracy of the model on the test images: 92.0 %
use 0.0min9.524372100830078s
Epoch [180/200], Step [100/391] Loss: 0.0039
Epoch [180/200], Step [200/391] Loss: 0.0032
Epoch [180/200], Step [300/391] Loss: 0.0015
Accuracy of the model on the test images: 92.14 %
use 0.0min9.443933963775635s
Epoch [181/200], Step [100/391] Loss: 0.0194
Epoch [181/200], Step [200/391] Loss: 0.0006
Epoch [181/200], Step [300/391] Loss: 0.0025
Accuracy of the model on the test images: 92.12 %
use 0.0min9.407834529876709s
Epoch [182/200], Step [100/391] Loss: 0.0028
Epoch [182/200], Step [200/391] Loss: 0.0202
Epoch [182/200], Step [300/391] Loss: 0.0006
Accuracy of the model on the test images: 92.09 %
use 0.0min9.312544584274292s
Epoch [183/200], Step [100/391] Loss: 0.0053
Epoch [183/200], Step [200/391] Loss: 0.0114
Epoch [183/200], Step [300/391] Loss: 0.0164
Accuracy of the model on the test images: 92.13 %
use 0.0min9.36332654953003s
Epoch [184/200], Step [100/391] Loss: 0.0029
Epoch [184/200], Step [200/391] Loss: 0.0006
Epoch [184/200], Step [300/391] Loss: 0.0099
Accuracy of the model on the test images: 92.0 %
use 0.0min9.633113145828247s
Epoch [185/200], Step [100/391] Loss: 0.0061
Epoch [185/200], Step [200/391] Loss: 0.0019
Epoch [185/200], Step [300/391] Loss: 0.0058
Accuracy of the model on the test images: 92.18 %
use 0.0min9.323748588562012s
Epoch [186/200], Step [100/391] Loss: 0.0059
Epoch [186/200], Step [200/391] Loss: 0.0032
Epoch [186/200], Step [300/391] Loss: 0.0057
Accuracy of the model on the test images: 91.99 %
use 0.0min9.359752655029297s
Epoch [187/200], Step [100/391] Loss: 0.0226
Epoch [187/200], Step [200/391] Loss: 0.0102
Epoch [187/200], Step [300/391] Loss: 0.0042
Accuracy of the model on the test images: 92.19 %
use 0.0min9.701231718063354s
Epoch [188/200], Step [100/391] Loss: 0.0025
Epoch [188/200], Step [200/391] Loss: 0.0073
Epoch [188/200], Step [300/391] Loss: 0.0024
Accuracy of the model on the test images: 92.14 %
use 0.0min9.389238119125366s
Epoch [189/200], Step [100/391] Loss: 0.0076
Epoch [189/200], Step [200/391] Loss: 0.0035
Epoch [189/200], Step [300/391] Loss: 0.0082
Accuracy of the model on the test images: 92.01 %
use 0.0min9.795827627182007s
Epoch [190/200], Step [100/391] Loss: 0.0061
Epoch [190/200], Step [200/391] Loss: 0.0013
Epoch [190/200], Step [300/391] Loss: 0.0043
Accuracy of the model on the test images: 92.23 %
use 0.0min9.78023076057434s
Epoch [191/200], Step [100/391] Loss: 0.0151
Epoch [191/200], Step [200/391] Loss: 0.0032
Epoch [191/200], Step [300/391] Loss: 0.0004
Accuracy of the model on the test images: 92.05 %
use 0.0min9.369535446166992s
Epoch [192/200], Step [100/391] Loss: 0.0025
Epoch [192/200], Step [200/391] Loss: 0.0217
Epoch [192/200], Step [300/391] Loss: 0.0016
Accuracy of the model on the test images: 91.99 %
use 0.0min9.30561113357544s
Epoch [193/200], Step [100/391] Loss: 0.0032
Epoch [193/200], Step [200/391] Loss: 0.0024
Epoch [193/200], Step [300/391] Loss: 0.0072
Accuracy of the model on the test images: 92.21 %
use 0.0min9.247185707092285s
Epoch [194/200], Step [100/391] Loss: 0.0148
Epoch [194/200], Step [200/391] Loss: 0.0200
Epoch [194/200], Step [300/391] Loss: 0.0037
Accuracy of the model on the test images: 92.12 %
use 0.0min9.572056770324707s
Epoch [195/200], Step [100/391] Loss: 0.0064
Epoch [195/200], Step [200/391] Loss: 0.0071
Epoch [195/200], Step [300/391] Loss: 0.0082
Accuracy of the model on the test images: 92.07 %
use 0.0min9.338852405548096s
Epoch [196/200], Step [100/391] Loss: 0.0036
Epoch [196/200], Step [200/391] Loss: 0.0075
Epoch [196/200], Step [300/391] Loss: 0.0093
Accuracy of the model on the test images: 92.06 %
use 0.0min9.353738069534302s
Epoch [197/200], Step [100/391] Loss: 0.0023
Epoch [197/200], Step [200/391] Loss: 0.0010
Epoch [197/200], Step [300/391] Loss: 0.0024
Accuracy of the model on the test images: 92.01 %
use 0.0min9.38978362083435s
Epoch [198/200], Step [100/391] Loss: 0.0038
Epoch [198/200], Step [200/391] Loss: 0.0065
Epoch [198/200], Step [300/391] Loss: 0.0092
Accuracy of the model on the test images: 92.18 %
use 0.0min9.295950651168823s
Epoch [199/200], Step [100/391] Loss: 0.0013
Epoch [199/200], Step [200/391] Loss: 0.0005
Epoch [199/200], Step [300/391] Loss: 0.0139
Accuracy of the model on the test images: 92.14 %
use 0.0min9.267261505126953s
Epoch [200/200], Step [100/391] Loss: 0.0044
Epoch [200/200], Step [200/391] Loss: 0.0025
Epoch [200/200], Step [300/391] Loss: 0.0016
Accuracy of the model on the test images: 92.19 %
use 0.0min9.337836980819702s
Accuracy of the model on the test images: 99.852 %