How to modify the loss function as Dice + CE loss?

[project-monai]

How to modify the loss function as Dice + CE loss?

[Nic-Ma]

Hi @GOZGOXGOYGO ,

I don't see some obvious error in your loss function, maybe you can compare all your program with:
https://github.com/Project-MONAI/tutorials/blob/master/modules/dynunet_tutorial.ipynb
And please make sure your software env is OK to run a regular PyTorch program.

Thanks.

[GOZGOXGOYGO]

Hi,

I think this error might be a bug. Actually, my code can run the monai code with officially provided dice loss. But when I change this part, the code cannot work now. It seems that in the tutorial code, the modified loss function is used with SupervisedTrainer in the following form. But why can't just apply it in this tutorial code?
https://github.com/Project-MONAI/tutorials/blob/master/2d_segmentation/torch/unet_training_dict.py

Thank you.

    trainer = monai.engines.SupervisedTrainer(
        device=device,
        max_epochs=max_epochs,
        train_data_loader=train_loader,
        network=net,
        optimizer=opt,
        loss_function=DiceCELoss(),
        inferer=get_inferer(),
        key_train_metric=None,
        train_handlers=train_handlers,
        amp=amp,
    )
    trainer.run()

[Nic-Ma]

Hi @GOZGOXGOYGO ,

Could you please help provide a simple program to reproduce your issue? I didn't find any issue at a first look.

Thanks.