num_workers=0,
num_rep=20,
prob_rotate=1,
max_angel=180,
sagittal_size=dis_model.sagittal_size,
transverse_size=dis_model.transverse_size,
k_nearest=dis_model.k_nearest,
sagittal_shift=dis_model.sagittal_shift,
pin_memory=False,
sampling_strategy=None)
# 设定验证参数
valid_evaluator = Evaluator(
dis_model,
valid_studies,
'../data/lumbar_train51_annotation.json',
num_rep=5,
max_dist=6,
)
# 每个batch要训练的步骤
step_per_batch = len(train_dataloader)
# 设置Adam优化器，以及学习率
optimizer = torch.optim.AdamW(dis_model.parameters(), lr=1e-5)
# 设置最大训练轮次
max_step = 30 * step_per_batch
# 训练
fit_result = torch_utils.fit(
dis_model,
train_data=train_dataloader,
valid_data=None,
batch_size=8,
num_workers=0,
num_rep=20,
prob_rotate=1,
max_angel=180,
sagittal_size=dis_model.sagittal_size,
transverse_size=dis_model.sagittal_size,
k_nearest=0,
max_dist=6,
sagittal_shift=1,
pin_memory=False)
# 设定验证参数
valid_evaluator = Evaluator(dis_model,
valid_studies,
'../data/lumbar_train51_annotation.json',
num_rep=20,
max_dist=6,
metric='key point recall')
# 每个batch要训练的步骤
step_per_batch = len(train_dataloader)
# 设置Adam优化器，以及学习率
optimizer = torch.optim.AdamW(dis_model.parameters(), lr=1e-5)
# 设置最大训练轮次
max_step = 50 * step_per_batch
# 训练
fit_result = torch_utils.fit(
dis_model,
train_data=train_dataloader,
valid_data=None,
optimizer=optimizer,