I'm very interested in exploring the application of LDM in virtual try-on and inspired by your work. But I'm confused by the second third row of Tab. 4 in your paper.
I notice the performance doesn't drop obviously with empty strings (row 1) or textual elements (row 2). How can I get textual elements? Maybe pass the garment images directly through VE to U-NET?
Moreover, why does the performance drop dramatically with f_theta? even much worse than empty strings?
Hi, thanks for sharing your great work!
I'm very interested in exploring the application of LDM in virtual try-on and inspired by your work. But I'm confused by the second third row of Tab. 4 in your paper.
I notice the performance doesn't drop obviously with empty strings (row 1) or textual elements (row 2). How can I get textual elements? Maybe pass the garment images directly through VE to U-NET?
Moreover, why does the performance drop dramatically with f_theta? even much worse than empty strings?
Looking forward to your reply! Thank you again!