questions about training on btcv dataset

Hello, I'm sorry to bother you. I'm a novice in medical image segmentation and encountered the following problem while trying to train the btcv dataset using your code:

First, I use label transfer. py to generate post label.h5 file, as the goal is to split the btcv dataset, I have made the following modifications, and the other settings remain unchanged: ORGAN_LIST = '../dataset/dataset_list/btcv_train.txt' NUM_WORKER = 8 NUM_CLASS = 13 TRANSFER_LIST = ['01'] TEMPLATE={ '01': [1,2,3,4,5,6,7,8,9,10,11,12,13], 2.Secondly, I attempted to use train.py to train the btcv dataset, and my settings are as follows: CUDA_VISIBLE_DEVICES=5 python train.py --phase train --data_root_path /data/ContinualLearning/dataset/ --train_data_txt_path /data/ContinualLearning/dataset/dataset_list/mybtcv_train.txt --val_data_txt_path /data/ContinualLearning/dataset/dataset_list/mybtcv_val.txt --organ_list 1 2 3 4 5 6 7 8 9 10 11 12 13 --max_epoch 1001 --warmup_epoch 15 --batch_size 1 --num_samples 1 --lr 1e-4 --model swinunetr --trans_encoding word_embedding --out_nonlinear softmax --out_channels 14 --log_name /data/ContinualLearning/log1018/ --pretrain /data/ContinualLearning/pretrained_weights/swin_unetr.base_5000ep_f48_lr2e-4_pretrained.pt At the same time, I made the following modifications to the num_classes of the two loss functions in train： `class DiceLoss(nn.Module): def init(self, weight=None, ignore_index=None, num_classes=14, kwargs): super(DiceLoss, self).init() self.kwargs = kwargs self.weight = weight self.ignore_index = ignore_index self.num_classes = num_classes self.dice = loss.BinaryDiceLoss(self.kwargs)

def forward(self, predict, target, organ_list): total_loss = [] predict = F.sigmoid(predict)
```
total_loss = []
B = predict.shape[0]

for b in range(B):
    for organ in organ_list:
        dice_loss = self.dice(predict[b, organ-1], target[b, organ-1])
        total_loss.append(dice_loss)

total_loss = torch.stack(total_loss)

return total_loss.sum()/total_loss.shape[0]
```

class Multi_BCELoss(nn.Module): def init(self, ignore_index=None, num_classes=14, **kwargs): super(Multi_BCELoss, self).init() self.kwargs = kwargs self.num_classes = num_classes self.ignore_index = ignore_index self.criterion = nn.BCEWithLogitsLoss()

def forward(self, predict, target, organ_list):
    assert predict.shape[2:] == target.shape[2:], 'predict & target shape do not match'
    total_loss = []
    B = predict.shape[0]

    for b in range(B):
        for organ in organ_list:
            ce_loss = self.criterion(predict[b, organ-1], target[b, organ-1])
            total_loss.append(ce_loss)
    total_loss = torch.stack(total_loss)

    return total_loss.sum()/total_loss.shape[0]`

3.The above are my modifications, but during the training process, the dice loss did not decrease and remained around 0.97. I speculate that there may be a problem with the category settings in my modifications.May I ask if there is a problem with my modifications? I'm sorry to bother you, and I hope to receive guidance from you or someone else who is interested in the code. Thank you!

MrGiovanni / ContinualLearning

questions about training on btcv dataset #10