Open dennislysenko opened 11 months ago
Hi dennislysenko, there is image generation method that can be conditioned on other signals like semantic segmentation. I don't think they're specifically trained for body shape, but eventually they will work well and for sure they can be finetuned.
Is it possible to go from image2image given a modified shape file? E.g. I put a photo of myself if I'm fat, regressor => creates a "fat" shape. I go S2A, modify some attributes to slim me and make me more muscular, A2S => create a new shape. Is there any research that would take the updated shape and the source image and generate a new image with that updated shape?