Describe the bug
In the implementation of the UNetR, specifically in the addition of the MLP Layer, the mlp_dim parameter is not used. Upon further inspection, it seems that this is a parameter to be used in the hidden layer of the MLP, but it is never passed to the function. Instead, the function by default uses the output parameters as the number of hidden parameters. It is actually incorrect according to the original paper and may change performance.
To Reproduce
Expected behavior
Screenshots
GaNDLF Version
Version information of the GaNDLF package in the virtual environment. : 0.0.16-dev
Desktop (please complete the following information):
Geeks-Sid
commented
1 year ago
Describe the bug In the implementation of the UNetR, specifically in the addition of the MLP Layer, the mlp_dim parameter is not used. Upon further inspection, it seems that this is a parameter to be used in the hidden layer of the MLP, but it is never passed to the function. Instead, the function by default uses the output parameters as the number of hidden parameters. It is actually incorrect according to the original paper and may change performance.
To Reproduce
Expected behavior
Screenshots
GaNDLF Version Version information of the GaNDLF package in the virtual environment. : 0.0.16-dev
Desktop (please complete the following information):
Additional context