for celeb hq images the output is fine. when i use my own photos the reconstruction is weird. #330
Description
i used this with pretrained model of styleinversion.it works fine with celebhq dataset ,but when i tried with my gallery images or celebrity images (from google) the reconstruction is abnormal. when i use deepika's photo the result is as below. how to get better reconstruction .as this model is trained with ffhq data,for reconstruction it is not given good results. even i tried e4e model. if i want to give training with my own dataset which type and resolution should be used to get good results .please provide information about creating own dataset for better preserving face.
