def eval(args, model, dataloader):
print('\nStart Test!')
for c in os.listdir(os.path.join("./checkpoints", args.net_work)):
pretrained_model_path = os.path.join("./checkpoints", args.net_work,
c) # 最后一个模型(最好的)
print("Load best model " + '\"' + os.path.abspath(pretrained_model_path) +
'\"')
checkpoint = torch.load(pretrained_model_path)
model.load_state_dict(checkpoint['state_dict'])
with torch.no_grad():
model.eval()
test_progressor = pb.Test_ProgressBar(total=len(dataloader),
model_name=args.net_work)
total_Dice = []
total_Acc = []
total_jaccard = []
total_Sensitivity = []
total_Specificity = []
for i, (data, [label, label_path]) in enumerate(dataloader):
test_progressor.current = i
if torch.cuda.is_available() and args.use_gpu:
data = data.cuda()
label = label.cuda()
# get RGB predict image
predict = model(data)
Dice, Acc, jaccard, Sensitivity, Specificity = u.eval_single_seg(
predict, label)
predict = torch.round(torch.sigmoid(predict)).byte()
predict_seg = predict.data.cpu().numpy().squeeze() * 255
img = Image.fromarray(predict_seg, mode='L')
label_name = label_path[0].split(os.sep)[-1].split('.')[0] + '.png'
img.save(args.result_path + label_name)
total_Dice += Dice
total_Acc += Acc
total_jaccard += jaccard
total_Sensitivity += Sensitivity
total_Specificity += Specificity
dice = sum(total_Dice) / len(total_Dice)
acc = sum(total_Acc) / len(total_Acc)
jac = sum(total_jaccard) / len(total_jaccard)
sen = sum(total_Sensitivity) / len(total_Sensitivity)
spe = sum(total_Specificity) / len(total_Specificity)
test_progressor.val = [dice, acc, jac, sen, spe]
test_progressor()
test_progressor.done()
def val(args, model, dataloader):
print('\n')
print('Start Validation!')
with torch.no_grad():
model.eval()
tbar = tqdm.tqdm(dataloader, desc='\r')
total_inter, total_union, total_correct, total_label = 0, 0, 0, 0
total_Dice = []
total_Acc = []
total_jaccard = []
total_Sensitivity = []
total_Specificity = []
for i, (data, label) in enumerate(tbar):
# tbar.update()
if torch.cuda.is_available() and args.use_gpu:
data = data.cuda()
label = label.cuda()
# get RGB predict image
aux_predict, predict = model(data)
Dice, Acc, jaccard, Sensitivity, Specificity = u.eval_single_seg(
predict, label)
total_Dice += Dice
total_Acc += Acc
total_jaccard += jaccard
total_Sensitivity += Sensitivity
total_Specificity += Specificity
dice = sum(total_Dice) / len(total_Dice)
acc = sum(total_Acc) / len(total_Acc)
jac = sum(total_jaccard) / len(total_jaccard)
sen = sum(total_Sensitivity) / len(total_Sensitivity)
spe = sum(total_Specificity) / len(total_Specificity)
tbar.set_description(
'Dice: %.3f, Acc: %.3f, Jac: %.3f, Sen: %.3f, Spe: %.3f' %
(dice, acc, jac, sen, spe))
print('Dice:', dice)
print('Acc:', acc)
print('Jac:', jac)
print('Sen:', sen)
print('Spe:', spe)
return dice, acc, jac, sen, spe
def eval(args, model, dataloader):
print('\nStart Test!')
for c in os.listdir(os.path.join("./checkpoints",args.net_work)):
pretrained_model_path = os.path.join("./checkpoints",args.net_work,c) # 最后一个模型(最好的)
break
print("Load best model "+'\"'+os.path.abspath(pretrained_model_path)+'\"')
checkpoint = torch.load(pretrained_model_path)
model.load_state_dict(checkpoint['state_dict'])
with torch.no_grad():
model.eval()
test_progressor = pb.Test_ProgressBar(total=len(dataloader),model_name=args.net_work)
total_Dice = []
total_Acc = []
total_jaccard = []
total_Sensitivity = []
total_Specificity = []
for i, (data, label) in enumerate(dataloader):
test_progressor.current = i
if torch.cuda.is_available() and args.use_gpu:
data = data.cuda()
label = label.cuda()
# get RGB predict image
predict = model(data)
Dice, Acc, jaccard, Sensitivity, Specificity = u.eval_single_seg(predict, label)
total_Dice += Dice
total_Acc += Acc
total_jaccard += jaccard
total_Sensitivity += Sensitivity
total_Specificity += Specificity
dice = sum(total_Dice) / len(total_Dice)
acc = sum(total_Acc) / len(total_Acc)
jac = sum(total_jaccard) / len(total_jaccard)
sen = sum(total_Sensitivity) / len(total_Sensitivity)
spe = sum(total_Specificity) / len(total_Specificity)
test_progressor.val=[dice,acc,jac,sen,spe]
test_progressor()
test_progressor.done()
def val(args, model, dataloader, k_fold, epoch):
with torch.no_grad():
model.eval()
Dice_m = u.AverageMeter()
Acc_m = u.AverageMeter()
jaccard_m = u.AverageMeter()
Sensitivity_m = u.AverageMeter()
Specificity_m = u.AverageMeter()
eval_progressor = pb.Val_ProgressBar(mode='val',
epoch=epoch + 1,
fold=k_fold,
model_name=args.net_work,
net_index=args.net_index,
total=len(dataloader))
for i, (data, label) in enumerate(dataloader):
eval_progressor.current = i
if torch.cuda.is_available() and args.use_gpu:
data = data.cuda()
label = label.cuda()
aux_predict, predict = model(data)
# 获取评价指标
Dice, Acc, jaccard, Sensitivity, Specificity = u.eval_single_seg(
predict, label)
Dice_m.update(Dice)
Acc_m.update(Acc)
jaccard_m.update(jaccard)
Sensitivity_m.update(Sensitivity)
Specificity_m.update(Specificity)
dice, acc, jac, sen, spe = Dice_m.avg, Acc_m.avg, jaccard_m.avg, Sensitivity_m.avg, Specificity_m.avg
eval_progressor.val = [dice, acc, jac, sen, spe]
# 更新进度条
eval_progressor()
eval_progressor.done()
# print('Dice:', dice)
# print('Acc:', acc)
# print('Jac:', jac)
# print('Sen:', sen)
# print('Spe:', spe)
return dice, acc, jac, sen, spe
def val(args, model, dataloader):
with torch.no_grad():
model.eval()
val_progressor = pb.Val_ProgressBar(
model_name=args.net_work, total=len(dataloader)) # 验证进度条，用于显示指标
total_Dice = []
total_Acc = []
total_jaccard = []
total_Sensitivity = []
total_Specificity = []
for i, (data, label) in enumerate(dataloader):
val_progressor.current = i
if torch.cuda.is_available() and args.use_gpu:
data = data.cuda()
label = label.cuda()
# get RGB predict image
predict = model(data)
Dice, Acc, jaccard, Sensitivity, Specificity = u.eval_single_seg(
predict, label) # 每个指标返回的是batchsize长的列表
# 将每个batch的列表相加，在迭代中动态显示指标的平均值（最终值）
total_Dice += Dice
total_Acc += Acc
total_jaccard += jaccard
total_Sensitivity += Sensitivity
total_Specificity += Specificity
# 表示对batchsize个值取平均，len=batchsize
dice = sum(total_Dice) / len(total_Dice)
acc = sum(total_Acc) / len(total_Acc)
jac = sum(total_jaccard) / len(total_jaccard)
sen = sum(total_Sensitivity) / len(total_Sensitivity)
spe = sum(total_Specificity) / len(total_Specificity)
val_progressor.val = [dice, acc, jac, sen, spe]
val_progressor()
val_progressor.done()
return dice, acc, jac, sen, spe