Details in the Table 2 of the Section 4.3 Quantitative Comparision

[LeCongThuong]

Dear author,
Thank for sharing your great work.

I want to ask you the detail of Table 2 of the Section 4.3 in the paper when you compare the models for the image-to-image tasks on edges2shoes, edge2handbags, faces2comics.

In my understanding, the training combines 2 phases:

1. Train VQGAN for both image domains (for instance, edge image domain, and shoes domain).
2. Freeze the pretrained VQGAN model from step 1, train LatentBrownianBridgeModel.

To get the result you report in the Table 2, you use only one VQGAN model to train combination of both image domains, or you use 2 VQGAN models, each for one image domain in the phase 1?