Question about the whole algorithm?

[Rashfu]

Is your proposed algorithm trained in the supervised way? I see that the input of STTR contains GT disparity and occlusion region during training phase.

[mli0603]

The provided code is supervised yes. Though you can train using proxy cues such as photometric loss. But it doesn't work as well as supervised.

[Rashfu]

OK, I get it. The GT disparity of Relative Response Loss paper is obtained by SFM automatically, so they call it self-supervised training scheme. But yours is provided by the dataset itself, so it's supervised. Am I right?

[Magicboomliu]

I think so . In Sttr is supevised