ytongxie / cotr Goto Github PK

Hey,

Thanks for publicizing your Code and the awesome architecture you designed :)
I was wondering if you have some checkpoints for public datasets you used that you already trained as I would like to reuse the original ones in a research project.

Best regards,
Tassilo

When I want to test, I follow the instructions in README.md, but it seems to start training again? What's going on here?

Hi,
I'm trying to train the CoTr model on Hippocampus dataset that I downloaded from http://medicaldecathlon.com/
However, I'm getting the error that the size of the tensors when performing the in skip connection operation doesn't match.

The error is basically somewhere in the CoTr/network_architecture/ResTranUnet.py line 156
x = x + skip2

I tried to solve the issue but I don't fully understand the source of the error.
Any help will be really appreciated

Thank you

作者，你好，可以解释下这段代码的意思吗？
def get_reference_points(spatial_shapes, valid_ratios, device):
reference_points_list = []
for lvl, (D_, H_, W_) in enumerate(spatial_shapes):

        ref_d, ref_y, ref_x = torch.meshgrid(torch.linspace(0.5, D_ - 0.5, D_, dtype=torch.float32, device=device),
                                             torch.linspace(0.5, H_ - 0.5, H_, dtype=torch.float32, device=device),
                                             torch.linspace(0.5, W_ - 0.5, W_, dtype=torch.float32, device=device))

        ref_d = ref_d.reshape(-1)[None] / (valid_ratios[:, None, lvl, 0] * D_)
        ref_y = ref_y.reshape(-1)[None] / (valid_ratios[:, None, lvl, 2] * H_)
        ref_x = ref_x.reshape(-1)[None] / (valid_ratios[:, None, lvl, 1] * W_)

        ref = torch.stack((ref_d, ref_x, ref_y), -1)   # D W H
        reference_points_list.append(ref)
    reference_points = torch.cat(reference_points_list, 1)
    reference_points = reference_points[:, :, None] * valid_ratios[:, None]
    return reference_points

I have some questions about the preprocessing of BTCV data. I preprocessed according to nnUNet's method, but running "run_trainng.py" shows: KeyError:'bcv_32'.