custom-diffusion icon indicating copy to clipboard operation
custom-diffusion copied to clipboard

Published 20 hours ago • verified publisher icon adobe-research

Unable to reproduce results

Open aisagarw opened this issue 2 years ago • 2 comments

I am using diffuser based multi-concept fine-tuning script for the case of cat and wooden_pot but I am not able to reproduce the kind of results shown in the image below.

image

For example- for the prompt: V1 cat sitting inside a V2 wooden pot and looking up, I get the following results on sampling post fine-tuning. image

I am using more than 2 GPUs while training. Are there any specific steps to follow that I might have missed?

aisagarw avatar Feb 08 '23 06:02 aisagarw

Hi, with the diffuser training script mentioned here for multiple concepts, I get the following results. new1 -cat-sitting-inside-a- new2 -wooden-pot-and

Our results in the paper were with the CompVis codebase, and there are minor differences between that and diffuser training. But I am able to get better composition with the diffuser training as well, compared to yours. Can you post the command that was used for training?

Thanks.

nupurkmr9 avatar Feb 09 '23 16:02 nupurkmr9

I have also encountered the same problem. Can you give me some guidance?

Details of the problem: https://github.com/adobe-research/custom-diffusion/issues/88

ZZZBBBZZZ avatar Jan 23 '24 08:01 ZZZBBBZZZ