Questions with 3D GAN inversion method

[binarydaddy]

Hello,
First of all, congrats on CVPR 2023 and thanks for sharing such wonderful work!
I had some unclear parts with 3D GAN inversion method you have described on the paper. (How you generated results in Figure 8.)

Is this "instance-selected domain adaptation combined with 3D GAN inversion" method simply inverting the modified image to EG3D W space and reconstructing images from it?

Or is the statement closer to JoJoGAN, where EG3D fine-tuning is included?

Thank you!