Hello, I see that in your ablation experiment, image editing tasks such as Textual Inversion and Dream Booth are compared. They modify the embedding or unet with a specific series of pictures to generate the expected picture. I would like to ask, for the tasks mentioned to generate emotional images, how can they be used to generate emotional images? Is there a cosine prepared data set to train the network to generate this anger themed image?
Hello, I see that in your ablation experiment, image editing tasks such as Textual Inversion and Dream Booth are compared. They modify the embedding or unet with a specific series of pictures to generate the expected picture. I would like to ask, for the tasks mentioned to generate emotional images, how can they be used to generate emotional images? Is there a cosine prepared data set to train the network to generate this anger themed image?