Exemplo n.º 1
0
def main(args):
    np.random.seed(args.seed)
    torch.manual_seed(args.seed)
    torch.cuda.manual_seed(args.seed)
    torch.cuda.manual_seed_all(args.seed)
    cudnn.benchmark = True
    torch.backends.cudnn.deterministic = True

    args.cuda = args.cuda and torch.cuda.is_available()
    if args.cuda:
        print('using cuda.')
        torch.set_default_tensor_type('torch.cuda.FloatTensor')
    else:
        torch.set_default_tensor_type('torch.FloatTensor')

    if not args.evaluate:
        # make symlink
        make_symlink_if_not_exists(osp.join(args.real_logs_dir, args.logs_dir),
                                   osp.dirname(osp.normpath(args.logs_dir)))

    # Save the args to disk
    if not args.evaluate:
        cfg_save_path = osp.join(args.logs_dir, 'cfg.txt')
        cfgs = vars(args)
        with open(cfg_save_path, 'w') as f:
            for k, v in cfgs.items():
                f.write('{}: {}\n'.format(k, v))

    # Create data loaders
    if args.height is None or args.width is None:
        args.height, args.width = (128, 128)

    if not args.evaluate:
        train_dataset, train_loader = \
          get_data(args.synthetic_train_data_dir, args.num_train,args.height, args.width, args.batch_size, args.workers, True)
    test_dataset, test_loader = \
      get_data(args.test_data_dir, args.num_test, args.height, args.width, args.batch_size, args.workers, False)

    # Create model
    if args.model_arch == 'resnet34':
        model = resnet.resnet34(num_classes=args.num_class)
        print('########## Using resnet34')

    elif args.model_arch == 'resnet18':
        model = resnet.resnet18(num_classes=args.num_class)
        print('########## Using resnet18')

    elif args.model_arch == 'mobilenet_v2':
        model = torchvision.models.mobilenet_v2(num_classes=args.num_class)
        print('########## Using mobilenet_v2')

    else:
        print('Wrong Model!')
        return

    # criterion = nn.CrossEntropyLoss()
    criterion = nn.MSELoss()

    # Load from checkpoint
    if args.evaluation_metric == 'accuracy':
        best_res = 0
    else:
        raise ValueError("Unsupported evaluation metric:",
                         args.evaluation_metric)

    start_epoch = 0
    start_iters = 0

    if args.resume:
        checkpoint = load_checkpoint(args.resume)
        model.load_state_dict(checkpoint['state_dict'])

        # compatibility with the epoch-wise evaluation version
        if 'epoch' in checkpoint.keys():
            # start_epoch = checkpoint['epoch']
            start_epoch = 0
        else:
            # start_iters = checkpoint['iters']
            start_epoch = int(start_iters //
                              len(train_loader)) if not args.evaluate else 0
            start_iters = 0
            start_epoch = 0
        best_res = 0
        print("=> Start iters {}  best res {:.1%}".format(
            start_iters, best_res))

    if args.cuda:
        device = torch.device("cuda")
        model = model.to(device)
        model = nn.DataParallel(model)

    # Evaluator
    evaluator = Evaluator(model, args.evaluation_metric, args.logs_dir,
                          criterion, args.cuda)

    if args.evaluate:
        print('Test on {0}:'.format(args.test_data_dir))
        if len(args.vis_dir) > 0:
            vis_dir = osp.join(args.logs_dir, args.vis_dir)
            if not osp.exists(vis_dir):
                os.makedirs(vis_dir)
        else:
            vis_dir = None

        start = time.time()
        evaluator.evaluate(test_loader, dataset=test_dataset, vis_dir=vis_dir)
        print('it took {0} s.'.format(time.time() - start))
        return

    # Optimizer
    param_groups = model.parameters()
    param_groups = filter(lambda p: p.requires_grad, param_groups)
    optimizer = optim.Adadelta(param_groups,
                               lr=args.lr,
                               weight_decay=args.weight_decay)
    scheduler = optim.lr_scheduler.MultiStepLR(optimizer,
                                               milestones=args.milestones,
                                               gamma=0.1)

    # Trainer
    loss_weights = {}
    loss_weights['loss_rec'] = 1.
    if args.debug:
        args.print_freq = 1
    trainer = Trainer(model,
                      args.evaluation_metric,
                      args.logs_dir,
                      criterion,
                      iters=start_iters,
                      best_res=best_res,
                      grad_clip=args.grad_clip,
                      use_cuda=args.cuda)

    # Start training
    evaluator.evaluate(test_loader, step=0, dataset=test_dataset)
    for epoch in range(start_epoch, args.epochs):
        print('here')
        scheduler.step(epoch)
        current_lr = optimizer.param_groups[0]['lr']
        trainer.train(epoch,
                      train_loader,
                      optimizer,
                      current_lr,
                      print_freq=args.print_freq,
                      is_debug=args.debug,
                      evaluator=evaluator,
                      test_loader=test_loader,
                      test_dataset=test_dataset,
                      test_freq=args.test_freq)

    # Final test
    print('Test with best model:')
    checkpoint = load_checkpoint(osp.join(args.logs_dir, 'model_best.pth.tar'))
    model.module.load_state_dict(checkpoint['state_dict'])
    evaluator.evaluate(test_loader, dataset=test_dataset)
Exemplo n.º 2
0
def main(args):
    np.random.seed(args.seed)
    torch.manual_seed(args.seed)
    torch.cuda.manual_seed(args.seed)
    torch.cuda.manual_seed_all(args.seed)
    cudnn.benchmark = True
    torch.backends.cudnn.deterministic = True

    args.cuda = args.cuda and torch.cuda.is_available()
    if args.cuda:
        print('using cuda.')
        torch.set_default_tensor_type('torch.cuda.FloatTensor')
    else:
        torch.set_default_tensor_type('torch.FloatTensor')

    # Redirect print to both console and log file
    if not args.evaluate:
        # make symlink
        if not os.path.exists(args.logs_dir):
            os.makedirs(args.logs_dir)
        make_symlink_if_not_exists(osp.join(args.real_logs_dir, args.logs_dir),
                                   osp.dirname(osp.normpath(args.logs_dir)))
        sys.stdout = Logger(osp.join(args.logs_dir, 'log.txt'))

        train_tfLogger = TFLogger(osp.join(args.logs_dir, 'train'))
        eval_tfLogger = TFLogger(osp.join(args.logs_dir, 'eval'))

    # Save the args to disk
    if not args.evaluate:
        cfg_save_path = osp.join(args.logs_dir, 'cfg.txt')
        cfgs = vars(args)
        with open(cfg_save_path, 'w') as f:
            for k, v in cfgs.items():
                f.write('{}: {}\n'.format(k, v))

    # Create data loaders
    if args.height is None or args.width is None:
        args.height, args.width = (32, 100)
    print('height:', args.height, ' width: ', args.width)

    if not args.evaluate:
        train_dataset, train_loader = \
          get_data(args.train_data_dir, args.voc_type, args.max_len, args.num_train,
                   args.height, args.width, args.batch_size, args.workers, True, args.keep_ratio, n_max_samples=args.n_max_samples)
    test_dataset, test_loader = \
      get_data(args.test_data_dir, args.voc_type, args.max_len, args.num_test,
               args.height, args.width, args.batch_size, args.workers, False, args.keep_ratio)

    if args.evaluate:
        max_len = test_dataset.max_len
    else:
        max_len = max(train_dataset.max_len, test_dataset.max_len)
        train_dataset.max_len = test_dataset.max_len = max_len
    # Create model
    model = ModelBuilder(arch=args.arch,
                         rec_num_classes=test_dataset.rec_num_classes,
                         sDim=args.decoder_sdim,
                         attDim=args.attDim,
                         max_len_labels=max_len,
                         eos=test_dataset.char2id[test_dataset.EOS],
                         args=args,
                         STN_ON=args.STN_ON)
    #print('model: ', model)
    # import ipdb; ipdb.set_trace()
    params_num = sum(p.numel() for p in model.parameters() if p.requires_grad)
    encoder_flops, _ = get_model_complexity_info(model.encoder,
                                                 input_res=(3, 32, 100),
                                                 as_strings=False)
    print('num of parameters: ', params_num)
    print('encoder flops: ', encoder_flops)

    # Load from checkpoint
    if args.evaluation_metric == 'accuracy':
        best_res = 0
    elif args.evaluation_metric == 'editdistance':
        best_res = math.inf
    else:
        raise ValueError("Unsupported evaluation metric:",
                         args.evaluation_metric)
    start_epoch = 0
    start_iters = 0
    if args.resume:
        checkpoint = load_checkpoint(args.resume)
        model.load_state_dict(checkpoint['state_dict'])

        # compatibility with the epoch-wise evaluation version
        if 'epoch' in checkpoint.keys():
            start_epoch = checkpoint['epoch']
        else:
            start_iters = checkpoint['iters']
            start_epoch = int(start_iters //
                              len(train_loader)) if not args.evaluate else 0
        best_res = checkpoint['best_res']
        print("=> Start iters {}  best res {:.1%}".format(
            start_iters, best_res))

    if args.cuda:
        device = torch.device("cuda")
        model = model.to(device)
        model = nn.DataParallel(model)

    # Evaluator
    evaluator = Evaluator(model, args.evaluation_metric, args.cuda)

    if args.evaluate:
        print('Test on {0}:'.format(args.test_data_dir))
        if len(args.vis_dir) > 0:
            vis_dir = osp.join(args.logs_dir, args.vis_dir)
            if not osp.exists(vis_dir):
                os.makedirs(vis_dir)
        else:
            vis_dir = None

        start = time.time()
        evaluator.evaluate(test_loader, dataset=test_dataset, vis_dir=vis_dir)
        print('it took {0} s.'.format(time.time() - start))
        return

    # Optimizer
    param_groups = model.parameters()
    param_groups = filter(lambda p: p.requires_grad, param_groups)
    optimizer = optim.Adadelta(param_groups,
                               lr=args.lr,
                               weight_decay=args.weight_decay)
    scheduler = optim.lr_scheduler.MultiStepLR(optimizer,
                                               milestones=eval(
                                                   args.milestones),
                                               gamma=0.1)

    # Trainer
    loss_weights = {}
    loss_weights['loss_rec'] = 1.
    if args.debug:
        args.print_freq = 1
    trainer = Trainer(model,
                      args.evaluation_metric,
                      args.logs_dir,
                      iters=start_iters,
                      best_res=best_res,
                      grad_clip=args.grad_clip,
                      use_cuda=args.cuda,
                      loss_weights=loss_weights)

    # Start training
    evaluator.evaluate(test_loader,
                       step=0,
                       tfLogger=eval_tfLogger,
                       dataset=test_dataset)
    for epoch in range(start_epoch, args.epochs):
        scheduler.step(epoch)
        current_lr = optimizer.param_groups[0]['lr']
        trainer.train(epoch,
                      train_loader,
                      optimizer,
                      current_lr,
                      print_freq=args.print_freq,
                      train_tfLogger=train_tfLogger,
                      is_debug=args.debug,
                      evaluator=evaluator,
                      test_loader=test_loader,
                      eval_tfLogger=eval_tfLogger,
                      test_dataset=test_dataset)

    # Final test
    print('Test with best model:')
    checkpoint = load_checkpoint(osp.join(args.logs_dir, 'model_best.pth.tar'))
    model.module.load_state_dict(checkpoint['state_dict'])
    evaluator.evaluate(test_loader, dataset=test_dataset)

    # Close the tensorboard logger
    train_tfLogger.close()
    eval_tfLogger.close()
Exemplo n.º 3
0
def main(args):
  np.random.seed(args.seed)
  torch.manual_seed(args.seed)
  torch.cuda.manual_seed(args.seed)
  torch.cuda.manual_seed_all(args.seed)
  cudnn.benchmark = True
  torch.backends.cudnn.deterministic = True

  args.cuda = args.cuda and torch.cuda.is_available()
  print(torch.cuda.is_available())
  if args.cuda:
    print('using cuda.')
    torch.set_default_tensor_type('torch.cuda.FloatTensor')
  else:
    torch.set_default_tensor_type('torch.FloatTensor')
  # Redirect print to both console and log file
  if not args.evaluate:
    # make symlink
    make_symlink_if_not_exists(osp.join(args.real_logs_dir, args.logs_dir), osp.dirname(osp.normpath(args.logs_dir)))
    sys.stdout = Logger(osp.join(args.logs_dir, 'log.txt'))
    train_tfLogger = TFLogger(osp.join(args.logs_dir, 'train'))
    eval_tfLogger = TFLogger(osp.join(args.logs_dir, 'eval'))

  # Save the args to disk
  if not args.evaluate:
    cfg_save_path = osp.join(args.logs_dir, 'cfg.txt')
    # print()
    cfgs = vars(args)
    with open(cfg_save_path, 'w') as f:
      for k, v in cfgs.items():
        f.write('{}: {}\n'.format(k, v))

  # Create data loaders
  if args.height is None or args.width is None:
    args.height, args.width = (32, 100)

  if not args.evaluate: 
    train_dataset, train_loader = \
      get_data(args.synthetic_train_data_dir, args.voc_type, args.max_len, args.num_train,
               args.height, args.width, args.batch_size, args.workers, True, args.keep_ratio)
    voc = get_vocabulary('ALLCASES_SYMBOLS', EOS='EOS', PADDING='PADDING', UNKNOWN='UNKNOWN')
    id2char = dict(zip(range(len(voc)), voc))
    char2number = dict(zip(voc, [0]*len(voc)))
    # for _, label, _ in train_dataset:
    #   # word = ''
    #   for i in label:
    #     if not id2char[i] in ['EOS','PADDING','UNKNOWN']:
    #       char2number[id2char[i]] += 1
    #       # word += id2char[i]
    # # print(char2number)
    # for key in char2number.keys():
    #   print("{}:{}".format(key, char2number[key]))
      
      

  test_dataset, test_loader = \
    get_data(args.test_data_dir, args.voc_type, args.max_len, args.num_test,
             args.height, args.width, args.batch_size, args.workers, False, args.keep_ratio)
  # print("len(trainset) ", len(train_dataset))

  if args.evaluate:
    max_len = test_dataset.max_len
  else:
    max_len = max(train_dataset.max_len, test_dataset.max_len)
    train_dataset.max_len = test_dataset.max_len = max_len
  # Create model
  

  model = ModelBuilder(arch=args.arch, rec_num_classes=test_dataset.rec_num_classes,
                       sDim=args.decoder_sdim, attDim=args.attDim, max_len_labels=max_len,
                       eos=test_dataset.char2id[test_dataset.EOS], STN_ON=args.STN_ON,
                       encoder_block= args.encoder_block, decoder_block= args.decoder_block)

  for param in model.decoder.parameters():
    if isinstance(param, Parameter):
      param.requires_grad = False

  # for param in model.encoder.parameters():
  #   param.requires_grad = False
  # for param in model.stn_head.parameters():
  #   param.requires_grad = False

  # Load from checkpoint
  if args.evaluation_metric == 'accuracy':
    best_res = 0
  elif args.evaluation_metric == 'editdistance':
    best_res = math.inf
  else:
    raise ValueError("Unsupported evaluation metric:", args.evaluation_metric)
  start_epoch = 0
  start_iters = 0
  if args.resume:
    print("args.resume: ",args.resume)
    checkpoint = load_checkpoint(args.resume)
    model.load_state_dict(checkpoint['state_dict'])
    # for param in model.stn_head.parameters():
    #   # print(param.data)
    #   param.requires_grad = False
    # for param in model.encoder.parameters():
    #   param.requires_grad = False

    # compatibility with the epoch-wise evaluation version
    if 'epoch' in checkpoint.keys():
      start_epoch = checkpoint['epoch']
    else:
      start_iters = checkpoint['iters']
      start_epoch = int(start_iters // len(train_loader)) if not args.evaluate else 0
    # checkpoint['best_res'] = 0.802
    best_res = checkpoint['best_res']
    print("=> Start iters {}  best res {:.1%}"
          .format(start_iters, best_res))
  
  if args.cuda:
    device = torch.device("cuda")
    model = model.to(device)
    model = nn.DataParallel(model)
  # Evaluator
  evaluator = Evaluator(model, args.evaluation_metric, args.cuda)

  if args.evaluate:
    print('Test on {0}:'.format(args.test_data_dir))
    if len(args.vis_dir) > 0:
      vis_dir = osp.join(args.logs_dir, args.vis_dir)
      if not osp.exists(vis_dir):
        os.makedirs(vis_dir)
    else:
      vis_dir = None

    start = time.time()
    # print(test_dataset.lexicons50)
    evaluator.evaluate(test_loader, dataset=test_dataset, vis_dir=vis_dir)
    print('it took {0} s.'.format(time.time() - start))
    return

  # Optimizer
  param_groups = model.parameters()
  # model.stn_head.weight.requires_grad = False
  # model.encoder.weight.requires_grad = False
  param_groups = filter(lambda p: p.requires_grad, param_groups)
  # optimizer = optim.Adadelta(param_groups, lr=args.lr, weight_decay=args.weight_decay)
  optimizer = optim.Adam(param_groups, lr=args.lr, betas=(0.9, 0.98), eps=1e-09, weight_decay=args.weight_decay, amsgrad=False)
  # optimizer = optim.SGD(param_groups, lr=args.lr, momentum=0.9)
  # optimizer = optim.AdamW(param_groups, lr=args.lr, betas=(0.9, 0.999), eps=1e-08, weight_decay=0.01, amsgrad=False)
  # optimizer = optim.ASGD(param_groups, lr=args.lr, lambd=0.0001, alpha=0.75, t0=1000000.0, weight_decay=0)
  # optimizer = optim.Adagrad(param_groups, lr=args.lr, lr_decay=0, weight_decay=0, initial_accumulator_value=0)
  scheduler = optim.lr_scheduler.CosineAnnealingLR(optimizer, len(train_loader))

  # Trainer
  loss_weights = {}
  loss_weights['loss_rec'] = 1.
  if args.debug:
    args.print_freq = 1
  trainer = Trainer(model, args.evaluation_metric, args.logs_dir, 
                    iters=start_iters, best_res=best_res, grad_clip=args.grad_clip,
                    use_cuda=args.cuda, loss_weights=loss_weights)

  # Start training
  # evaluator.evaluate(test_loader, step=0, tfLogger=eval_tfLogger, dataset=test_dataset)
  # print("args.epoch: ", args.epochs)
  for epoch in range(start_epoch, args.epochs):
    scheduler.step(epoch)
    current_lr = optimizer.param_groups[0]['lr']
    # current_lr = (1.0/(512.0**0.5))*min(1.0/float(trainer.iters + 1)**0.5, float(trainer.iters+1)*1.0/16000.0**1.5)
    # optimizer.param_groups[0]['lr'] = current_lr 
    trainer.train(epoch, train_loader, optimizer, current_lr,
                  print_freq=args.print_freq,
                  train_tfLogger=train_tfLogger, 
                  is_debug=args.debug,
                  evaluator=evaluator, 
                  test_loader=test_loader, 
                  eval_tfLogger=eval_tfLogger,
                  test_dataset=test_dataset)

  # Final test
  print('Test with best model:')
  checkpoint = load_checkpoint(osp.join(args.logs_dir, 'model_best.pth.tar'))
  model.load_state_dict(checkpoint['state_dict'])
  # print("naruto")
  evaluator.evaluate(test_loader, dataset=test_dataset)
  # print("sasuke")

  # Close the tensorboard logger
  train_tfLogger.close()
  eval_tfLogger.close()