Closed jaymefosa closed 9 months ago
Perhaps one reason is that densepose restricts the boundaries of a face/figure thereby distorting the original image. E.g., Reference image of a skinny person + Dense pose of a normal person will distort the skinny person
Two papers report better results with the 'thicker' style of conditioning images:
https://arxiv.org/pdf/2308.03610.pdf (avatarverse) https://arxiv.org/pdf/2311.16498.pdf (magicanimate)
Any reason you chose DWpose?