KeyError: 'text_only'
Thanks for your great work! I tried to train CustomNet on the sample dataset and got a KeyError of 'text_only'. How can I solve this issue?
https://github.com/TencentARC/CustomNet/blob/23a378fc8cfcec72181cc4fd625c708f9485e490/ldm/models/diffusion/ddpm.py#L756
Also, when I try to train customnet_inpaint, I cannot find “bg_concat”. https://github.com/TencentARC/CustomNet/blob/23a378fc8cfcec72181cc4fd625c708f9485e490/ldm/models/diffusion/ddpm.py#L772
I'm facing this issue too. Have you solved this?
Is it used to indicate whether there is only text and no background picture with 1 and 0 ?
I try to replace it with
text_only = (torch.rand(x.size(0), device=x.device) < uncond).float().unsqueeze(-1)
I dont know whether this is appropriate, but this works.