Are the original images generated from LlamaGen and then you compare the generated image with the editted image ? Other diffusion-based editting methods are performed based on the generated image?
Or the original images is directly acquired from benchmark testset ?
If it is the first case, the generated image and the editted image should be naturally similar, because you use the same model.
Are the original images generated from LlamaGen and then you compare the generated image with the editted image ? Other diffusion-based editting methods are performed based on the generated image?
Or the original images is directly acquired from benchmark testset ?
If it is the first case, the generated image and the editted image should be naturally similar, because you use the same model.