Parameters

[bf2harven]

First, Thanks for the great implementation. It really helped me to understand and play with segmentation by diffusion.

I would like to contribute pretrained models on Brats2020 and compare the performance with the original paper. To do so I would like to set the hyperparameters to be the same.

In the original implementation (https://github.com/WuJunde/MedSegDiff) they use the following parameters:
--num_res_blocks 2
--num_heads 1
--learn_sigma True
--attention_resolutions 16

Could you please confirm those parameters are equivalent to the following parameters in this code?:
(From the unet Class)

resnet_block_groups=8
attn_heads=4
attn_dim_head=32
full_self_attn ?

Thanks