How to generate the corresponding sketch img?

[zengjie617789]

Fllowing the demo, the result got is perfect. But when I want to use the costom clothes it failed. After check the code, I think it should generate the corresponding sketch img to match the model pose. As like the paper said:

While for the paired set we can directly use the human parsing mask to extract the garment of interest worn by the model and then feed it to the edge detection network, for the unpaired set we need to first create a warped version of the in-shop garment matching the body pose and shape of the target model.Following virtual try-on methods [50,55], we train a geometric transformation module that performs a thin-plate spline transformation [38] of the input garment and then refines the warped result using a U-Net model [40] like the sketch img and img below:

As far as I understand, does it mean the work only work under some certain condion?