scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(my_optim,
mode='min',
factor=0.1,
patience=5,
verbose=True)
loss = torch.nn.MSELoss()
# 读取数据集
train_dataset = ImagePairDataset_y(INPUTS_FOLDER_TRAIN, LABELS_FOLDER_TRAIN)
test_dataset = ImagePairDataset_y(INPUTS_FOLDER_TEST, LABELS_FOLDER_TEST)
train_iter = DataLoader(train_dataset, batch_size, shuffle=True)
test_iter = DataLoader(test_dataset, 1, shuffle=True)
print('Datasets loaded!')
# 训练
train(train_iter, test_iter, net, loss, my_optim, num_epochs, scheduler)
# 测试
print('Full test loss %.4f' % eval(test_iter, net, loss, 0))
# 保存网络
torch.save(net.state_dict(), SAVE_PATH)
print('Network saved: ' + SAVE_PATH)
# 读取网络
# net.load_state_dict(torch.load(LOAD_PATH))
print('Network loaded: ' + LOAD_PATH)
# 应用完整图片并写入
IMG_NAME = r'img_0'
IMG_OUT = OUTPUTS_FOLDER+r'/'+IMG_NAME+r'_' + \
patience=5,
verbose=True)
loss = torch.nn.MSELoss()
# 读取数据集
train_dataset = ImagePairDataset_y(INPUTS_FOLDER_TRAIN, LABELS_FOLDER_TRAIN)
test_dataset = ImagePairDataset_y(INPUTS_FOLDER_TEST, LABELS_FOLDER_TEST)
train_iter = DataLoader(train_dataset, batch_size, shuffle=True)
test_iter = DataLoader(test_dataset, 1, shuffle=True)
print('Datasets loaded!')
# 训练
train(train_iter,
test_iter,
net,
loss,
my_optim,
num_epochs,
scheduler,
need_gclip=True)
# 测试
print('Full test loss %.4f' % eval(test_iter, net, loss, 0))
# 保存网络
torch.save(net.state_dict(), SAVE_PATH)
print('Network saved: ' + SAVE_PATH)
# 读取网络
# net.load_state_dict(torch.load(LOAD_PATH))
print('Network loaded: ' + LOAD_PATH)