How to use the tensorboard.SummaryWriter function in tensorboard

# Use GPUs if available.
    if torch.cuda.is_available():
        model.cuda()
        model = torch.nn.DataParallel(model, device_ids=range(torch.cuda.device_count()))
        cudnn.benchmark = True

    # Define loss function and optimizer.
    criterion = nn.CrossEntropyLoss()
    optimizer = optim.SGD(model.parameters(),
                          lr=args.lr,
                          momentum=args.momentum,
                          nesterov=args.nesterov,
                          weight_decay=args.weight_decay)

    log_dir = 'logs/' + datetime.now().strftime('%B%d  %H:%M:%S')
    train_writer = SummaryWriter(os.path.join(log_dir ,'train'))
    test_writer = SummaryWriter(os.path.join(log_dir ,'test'))

    # Save argparse commandline to a file.
    with open(os.path.join(log_dir, 'commandline_args.txt'), 'w') as f:
        f.write('\n'.join(sys.argv[1:]))

    best_acc = 0  # best test accuracy

    for epoch in range(start_epoch, args.epochs):
        # Learning rate schedule.
        lr = adjust_learning_rate(optimizer, epoch + 1)
        train_writer.add_scalar('lr', lr, epoch)

        # Train for one epoch.
        train(train_loader, model, criterion, optimizer, train_writer, epoch)

from torch.utils.data import DataLoader

import utils.simul_transforms as simul_transforms
import utils.transforms as expanded_transforms
from config import ckpt_path
from datasets.cityscapes import CityScapes
from datasets.cityscapes.config import num_classes, ignored_label
from datasets.cityscapes.utils import colorize_mask
from models import FCN8ResNet
from utils.io import rmrf_mkdir
from utils.loss import CrossEntropyLoss2d
from utils.training import calculate_mean_iu

cudnn.benchmark = True
exp_name = 'fcn8resnet_cityscapes224*448'
writer = SummaryWriter('exp/' + exp_name)
pil_to_tensor = standard_transforms.ToTensor()
train_record = {'best_val_loss': 1e20, 'corr_mean_iu': 0, 'corr_epoch': -1}

train_args = {
    'batch_size': 16,
    'epoch_num': 800,  # I stop training only when val loss doesn't seem to decrease anymore, so just set a large value.
    'pretrained_lr': 1e-6,  # used for the pretrained layers of model
    'new_lr': 1e-6,  # used for the newly added layers of model
    'weight_decay': 5e-4,
    'snapshot': 'epoch_184_loss_0.8953_mean_iu_0.3923_lr_0.00001000.pth',  # empty string denotes initial training, otherwise it should be a string of snapshot name
    'print_freq': 50,
    'input_size': (224, 448),  # (height, width)
}

val_args = {
    'batch_size': 8,

from torch.autograd import Variable
from torch.backends import cudnn
from torch.optim.lr_scheduler import ReduceLROnPlateau
from torch.utils.data import DataLoader

import utils.joint_transforms as joint_transforms
import utils.transforms as extended_transforms
from datasets import cityscapes
from models import *
from utils import check_mkdir, evaluate, AverageMeter, CrossEntropyLoss2d

cudnn.benchmark = True

ckpt_path = '../../ckpt'
exp_name = 'cityscapes-fcn8s (caffe vgg)'
writer = SummaryWriter(os.path.join(ckpt_path, 'exp', exp_name))

args = {
    'train_batch_size': 12,
    'epoch_num': 500,
    'lr': 1e-10,
    'weight_decay': 5e-4,
    'input_size': (256, 512),
    'momentum': 0.99,
    'lr_patience': 100,  # large patience denotes fixed lr
    'snapshot': '',  # empty string denotes no snapshot
    'print_freq': 20,
    'val_batch_size': 16,
    'val_save_to_img_file': False,
    'val_img_sample_rate': 0.05  # randomly sample some validation results to display
}

def __init__(self, logging_dir=None, prefix='val', roc_path=None, class_names=None):
        self.prefix = prefix
        self.roc_path = roc_path
        self.class_names = class_names
        try:
            from tensorboard import SummaryWriter
            self.summary_writer = SummaryWriter(logging_dir)
        except ImportError:
            logging.error('You can install tensorboard via `pip install tensorboard`.')

def __init__(self, logging_dir, prefix=None, layers_list=None):
        self.prefix = prefix
        self.layers_list = layers_list
        try:
            from tensorboard import SummaryWriter
            self.summary_writer = SummaryWriter(logging_dir)
        except ImportError:
            logging.error('You can install tensorboard via `pip install tensorboard`.')

if torch.cuda.is_available():
        model.cuda()
        model = torch.nn.DataParallel(model, device_ids=range(torch.cuda.device_count()))
        cudnn.benchmark = True

    # Define loss function and optimizer.
    criterion = nn.CrossEntropyLoss()
    optimizer = optim.SGD(model.parameters(),
                          lr=args.lr,
                          momentum=args.momentum,
                          nesterov=args.nesterov,
                          weight_decay=args.weight_decay)

    log_dir = 'logs/' + datetime.now().strftime('%B%d  %H:%M:%S')
    train_writer = SummaryWriter(os.path.join(log_dir ,'train'))
    test_writer = SummaryWriter(os.path.join(log_dir ,'test'))

    # Save argparse commandline to a file.
    with open(os.path.join(log_dir, 'commandline_args.txt'), 'w') as f:
        f.write('\n'.join(sys.argv[1:]))

    best_acc = 0  # best test accuracy

    for epoch in range(start_epoch, args.epochs):
        # Learning rate schedule.
        lr = adjust_learning_rate(optimizer, epoch + 1)
        train_writer.add_scalar('lr', lr, epoch)


        # Train for one epoch.
        train(train_loader, model, criterion, optimizer, train_writer, epoch)

parser.add_argument('--imgsize'  ,type=int, default = 256, help='image size')
    parser.add_argument('--batchsize'  ,type=int, default = 20, help='batchsize')
    parser.add_argument('--workers'  ,type=int, default = 6, help='number of workers')
    parser.add_argument('--nepoch'  ,type=int, default = 50, help='number of epochs')
    parser.add_argument('--lr', type=float, default=0.002, help='learning rate, default=0.002')
    parser.add_argument('--beta1', type=float, default=0.5, help='beta1 for adam. default=0.5')
    parser.add_argument('--outf', type=str, default="filler_pano_pc", help='output folder')
    parser.add_argument('--model', type=str, default="", help='model path')
    parser.add_argument('--cepoch'  ,type=int, default = 0, help='current epoch')

    mean = torch.from_numpy(np.array([ 0.45725039,  0.44777581,  0.4146058 ]).astype(np.float32))

    opt = parser.parse_args()
    print(opt)

    writer = SummaryWriter(opt.outf + '/runs/'+datetime.now().strftime('%B%d  %H:%M:%S'))

    try:
        os.makedirs(opt.outf)
    except OSError:
        pass

    tf = transforms.Compose([
        transforms.Scale(opt.imgsize, opt.imgsize * 2),
        transforms.ToTensor(),
    ])
    
    mist_tf = transforms.Compose([
        transforms.ToTensor(),
    ])

parser.add_argument('--beta1', type=float, default=0.5, help='beta1 for adam. default=0.5')
    parser.add_argument('--outf', type=str, default="filler_pano_pc_full", help='output folder')
    parser.add_argument('--model', type=str, default="", help='model path')
    parser.add_argument('--cepoch', type=int, default = 0, help='current epoch')
    parser.add_argument('--loss', type=str, default="perceptual", help='l1 only')
    parser.add_argument('--init', type=str, default = "iden", help='init method')
    parser.add_argument('--l1', type=float, default = 0, help='add l1 loss')
    parser.add_argument('--color_coeff', type=float, default = 0, help='add color match loss')
    parser.add_argument('--cascade'  , action='store_true', help='debug mode')
    
    
    
    mean = torch.from_numpy(np.array([0.57441127,  0.54226291,  0.50356019]).astype(np.float32))
    opt = parser.parse_args()
    print(opt)
    writer = SummaryWriter(opt.outf + '/runs/'+datetime.now().strftime('%B%d  %H:%M:%S'))
    try:
        os.makedirs(opt.outf)
    except OSError:
        pass

    tf = transforms.Compose([
        transforms.ToTensor(),
    ])
    
    mist_tf = transforms.Compose([
        transforms.ToTensor(),
    ])
    
    d = PairDataset(root = opt.dataroot, transform=tf, mist_transform = mist_tf)
    d_test = PairDataset(root = opt.dataroot, transform=tf, mist_transform = mist_tf, train = False)

import torchvision.transforms as standard_transforms
import torchvision.utils as vutils
from tensorboard import SummaryWriter
from torch import optim
from torch.autograd import Variable
from torch.utils.data import DataLoader

import utils.joint_transforms as joint_transforms
import utils.transforms as extended_transforms
from datasets import voc
from models import *
from utils import check_mkdir, evaluate, AverageMeter, CrossEntropyLoss2d

ckpt_path = '../../ckpt'
exp_name = 'voc-psp_net'
writer = SummaryWriter(os.path.join(ckpt_path, 'exp', exp_name))

args = {
    'train_batch_size': 1,
    'lr': 1e-2 / sqrt(16 / 4),
    'lr_decay': 0.9,
    'max_iter': 3e4,
    'longer_size': 512,
    'crop_size': 473,
    'stride_rate': 2 / 3.,
    'weight_decay': 1e-4,
    'momentum': 0.9,
    'snapshot': '',
    'print_freq': 10,
    'val_save_to_img_file': True,
    'val_img_sample_rate': 0.01,  # randomly sample some validation results to display,
    'val_img_display_size': 384,

parser.add_argument('--imgsize'  ,type=int, default = 256, help='image size')
    parser.add_argument('--batchsize'  ,type=int, default = 36, help='batchsize')
    parser.add_argument('--workers'  ,type=int, default = 6, help='number of workers')
    parser.add_argument('--nepoch'  ,type=int, default = 50, help='number of epochs')
    parser.add_argument('--lr', type=float, default=0.002, help='learning rate, default=0.002')
    parser.add_argument('--beta1', type=float, default=0.5, help='beta1 for adam. default=0.5')
    parser.add_argument('--outf', type=str, default="filler", help='output folder')
    parser.add_argument('--model', type=str, default="", help='model path')


    mean = torch.from_numpy(np.array([ 0.45725039,  0.44777581,  0.4146058 ]).astype(np.float32))
    
    opt = parser.parse_args()
    print(opt)
    
    writer = SummaryWriter(opt.outf + '/runs/'+datetime.now().strftime('%B%d  %H:%M:%S'))
    
    try:
        os.makedirs(opt.outf)
    except OSError:
        pass


    if opt.debug:
        d = Places365Dataset(root = opt.dataroot, transform=transforms.Compose([
                                   vision_utils.RandomScale(opt.imgsize, int(opt.imgsize * 1.5)),
                                   transforms.RandomCrop(opt.imgsize),
                                   transforms.ToTensor(),
                               ]), train = False)
    else:
        d = Places365Dataset(root = opt.dataroot, transform=transforms.Compose([
                                   vision_utils.RandomScale(opt.imgsize, int(opt.imgsize * 1.5)),