Closed urimerhav closed 2 years ago
Hi,
Our code technically allows the flexibility in the lambda terms for the global / entire branch (for L_structure and L_appearance). In our experiments we set them to be equal. So for ablation, we set lambda_global_ssim = lambda_entire_ssim = 0
.
In some pairs the fidelity to the structure may be good even without it, but in many others it is crucial for maintaining structure (e.g., as in Fig. 9, in AFHQ, etc.)
Thanks for the clarification!
I'm trying to recreate the results in figure 9 in your paper. Specifically, disabling L_structure. Are both
lambda_global_ssim
andlambda_entire_ssim
taken together form the entirety of L_struct or am I misreading the code?So far tinkering with the various lambda terms, it seems the result always retains very high structural fidelity to the structure image.