search

Kid-Liet / Reg-GAN

149 stars 21 forks source link

WHY Registration after translation? #25

Open HuangShiqi128 opened 1 year ago

HuangShiqi128 commented 1 year ago

Dear Author,

I have a question relating your proposed framework when tackling misaligned data.

You proposed to introduce the registration network after translating the image, why you suggest this way? I think misaligned image can be firstly registered to provide aligned data and then perform aligned I2I translation.

I look forward to your answer.

Kid-Liet commented 1 year ago

Hello, if the registration network is first, it means it is a multi-modal registration network, and if registration is second, it is a single-modal registration. I think single-modal is relatively easier to train. Another point is that if the registration network is first, Adv loss will apply to the registration network, making things a bit complicated. But overall, I think the difference between the two should not be significant. The viewpoint is for reference only.

At 2023-04-22 00:43:57, "HuangShiqi128" @.***> wrote:

Dear Author,

I have a question relating your proposed framework when tackling misaligned data.

You proposed to introduce the registration network after translating the image, why you suggest this way? I think misaligned image can be firstly registered to provide aligned data and then perform aligned I2I translation.

I look forward to your answer.

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