search

Project-MONAI / research-contributions

Implementations of recent research prototypes/demonstrations using MONAI.
https://monai.io/
Apache License 2.0
1.02k stars 334 forks source link

Parameters of the problem #28

Closed jianglanlele closed 2 years ago

jianglanlele commented 2 years ago

Hi, while I set pos_embed: str = "conv", should I need to change conv_block: bool = False into conv_block: bool = True?

ahatamiz commented 2 years ago

Hi @jiangllll

Thanks for your comment. These indeed refer to different properties of the network. First, pos_embed denotes how the initial emebeddings are generated from the input images. One way to do it is to break the image into non-overlapping patches and use a perceptron layer to project into an embedding space. The other method is to just use a convolutional layer. Both techniques are equivalent but the implementation is different. We have provided both options by setting the pos_embed to conv or perceptron.

On the other hand, conv_block denotes whether or not to use convolutional blocks during the decoding process. By setting it true, these convolutional blocks are used in UnetrUpBlock after the transposed convolutional layers.

We have provided the settings that were originally used to train UNETR as the default parameters.

Thanks

jianglanlele commented 2 years ago

@ahatamiz Thanks for your reply，The explanation is very clear. Thanks