Have you tried using a semantic map instead of a probability map?

[vhank]

A semantic map,I mean,is the output of segmentation instead of the probability.I wonder if you tried using a semantic map and if you did,what was the result?Thank you.

[vhank]

When we use different segmentation models,the probability maps we get may be different but the final semantic maps are similar.So I wonder if the result doesn't depend on what segmentation model we use?

[Xintao]

1. We have tried it before. Using probability maps gives a better visual quality because it provides a delicate feature transform under the segmentation probability. See Sec 4.3 in the paper for more details.
2. If using the final semantic maps, the results do not depend on the segmentation model you use. But with probability maps, it may have influences.

[vhank]

Thanks for your kind reply!Could you tell me how I can train a SFTGAN model with x2 or x3 upscaling factor?How can I pre-train a model?

[Xintao]

I have not tried the x2 or x3 model. I think:

1. fine-tune the segmentation model.
2. fine-tuning the SFTGAN model from the X4 model is much easier.