One question about model design #1

Open
opened 2025-10-14 15:37:57 -06:00 by navan · 0 comments
Owner

Originally created by @holimario on 8/27/2025

Your work is very interesting, and thank you for releasing the code.
I have a small question regarding the model design. From what I see, your architecture seems closer to UNIT, in that it introduces a strategy for enforcing latent space consistency. But this also requires adding extra architectural components and more loss terms. I also noticed that your paper mentions CycleGAN as a reference—have you tried adopting a simpler CycleGAN-like design? It seems such a design might make training more stable. If you did try but eventually abandoned it, what was the reason?

*Originally created by @holimario on 8/27/2025* Your work is very interesting, and thank you for releasing the code. I have a small question regarding the model design. From what I see, your architecture seems closer to UNIT, in that it introduces a strategy for enforcing latent space consistency. But this also requires adding extra architectural components and more loss terms. I also noticed that your paper mentions CycleGAN as a reference—have you tried adopting a simpler CycleGAN-like design? It seems such a design might make training more stable. If you did try but eventually abandoned it, what was the reason?
Sign in to join this conversation.
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference: github/vec2vec#1
No description provided.