search

microsoft / Recursive-Cascaded-Networks

[ICCV 2019] Recursive Cascaded Networks for Unsupervised Medical Image Registration
https://arxiv.org/abs/1907.12353
MIT License
362 stars 87 forks source link

do you have shown results of deformable registration excluding affine network in your paper? #4

Closed Dongshengjiang closed 4 years ago

Dongshengjiang commented 4 years ago

hello, I am confused by the word in your ICCV paper:"Our 3-cascade VTN (in Table 2) already outperforms VTN (ADDD) (in Table 1) although they have similar network architectures". Is that mean the Table 2 3-cascade VTN is the results of only use recursive deformable net excluding affine net?

zsyzzsoft commented 4 years ago

The result is including the affine cascade. The only difference is that the similarity loss is only drawn on the last warped image, in contrast to VTN (ADDD).

Dongshengjiang commented 4 years ago

Thanks for your timely reply. By the way, does the preprocess image is not rigid aligned? traditionally, if the image is rigid aligned, and you add some B-spline deformations, then only deformable registration is necessary.

zsyzzsoft commented 4 years ago

We did not preprocess the images to be rigidly aligned (which should be very slow using traditional methods). VTN proves that the end-to-end affine network performs even better.