search

cwmok / DIRAC

This is the official Pytorch implementation of "Unsupervised Deformable Image Registration with Absent Correspondences in Pre-operative and Post-Recurrence Brain Tumor MRI Scans" (MICCAI 2022), written by Tony C. W. Mok and Albert C. S. Chung.
MIT License
36 stars 1 forks source link

normalized the deformation field #14

Closed 18370029656 closed 8 months ago

18370029656 commented 10 months ago

Hello, why do we need to normalize the deformation field flow and control the deformation variable to [-1,1], so that the maximum deformation does not exceed 1. Is this not very reasonable?

cwmok commented 10 months ago

Hi @18370029656,

This is inspired by ICNet. Basically, we want to restrict the maximum deformation potential of the network with the range_flow, since we focused on local non-linear deformation.

18370029656 commented 10 months ago

Hello, may I ask if there is anything I need to pay attention to when applying your idea to other frameworks? The regions without corresponding relationships that I calculated are always at the edges, as shown in the figure. )4L $R XMH2X4GVC 0V0E6K

cwmok commented 10 months ago

You should figure out the best combination of thresh_fw, thresh_bw and reg_code .

thresh_fw, thresh_bw: threshold of the missing correspondence regions for the forward and backward registration. (larger threshold = less sensitive to small missing correspondence regions.) reg_code: the weight of the smoothness regularization. (higher weight = smoother deformation field, less deformation potential.)

18370029656 commented 10 months ago

thanks for your quick reply！

---Original--- From: "Tony @.> Date: Wed, Nov 22, 2023 16:12 PM To: @.>; Cc: @.**@.>; Subject: Re: [cwmok/DIRAC] normalized the deformation field (Issue #14)

You should figure out the best combination of thresh_fw, thresh_bw and reg_code .

thresh_fw, thresh_bw: threshold of the missing correspondence regions for the forward and backward registration. (larger threshold = less sensitive to small missing correspondence regions.) reg_code: the weight of the smoothness regularization. (higher weight = smoother deformation field, less deformation potential.)

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>