A question about sample_per_volume and resize in promise2012 demo

[niftk]

A question about sample_per_volume and resize in promise2012 demo

[wyli]

Hi @EdwardMa12593 Yes, sample_per_volume parameter is actually ignored; that is, given one volume input, the sampler would always generate one window (by resizing the input).

The sampling strategy and window size are free system parameters, a uniform sampling of 64x64x64 windows could be another option... if it turns out to be a better choice for segmentation please do consider submitting the model to the model zoo :)

For the final problem, it seems the mismatch is caused by the anisotropic resizing, could you please elaborate the steps to replicate the results? many thanks!

[JunMa11]

Dear @wyli
Thanks for your reply.
Is there a guideline about how to submit the model to the model zoo?
I'm not allowed to register CMICLab.

[JunMa11]

For the mismatch which is caused by the anisotropic resizing, I followed the python code in the Notebook to generate the results.
The inference parameters are

[INFERENCE]
border = (5, 5, 5)
#inference_iter = 10
save_seg_dir = output/
output_interp_order = 0
spatial_window_size = (0, 0, 3)

The documentation on readthedocs.io helps a lot to me.
However, I am still confused about why settint the spatial_window_size = (0, 0, 3)?
Is this the reasion for the mismatch?