AlbedoGAN generate bad results

[XuankeShi]

Pretrained weights won't give you results like the demo you showed, or even bad ones. I have no doubts about your work.

[XuankeShi]

I hope you can release the GAN inversion part of the code to ensure the texture generation is robust. What do u think of it.

[aashishrai3799]

While you have the freedom to make your own decisions, it's worth noting that we already possess ample results showcasing the effectiveness of our work on standard benchmarks such as Now and REALY3D, along with qualitative evidence. Admittedly, there are inconsistencies in the inference code, but these can be easily addressed with minor engineering adjustments. The Demo Results I provided were generated using the same inference code. Unfortunately, since the project has been discontinued for around 8 months, I've been pressed for time to address these issues, such as incorporating cv2.cvtColor into the inference code. I'll still try my best to solve some of these issues soon after the ECCV deadline to match with the paper results.