KeyError: 'text_only'

[rootpine]

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?

CustomNet/ldm/models/diffusion/ddpm.py

Line 756 in 23a378f

text_only = batch["text_only"] # (B,1)

[rootpine]

Also, when I try to train customnet_inpaint, I cannot find “bg_concat”.

CustomNet/ldm/models/diffusion/ddpm.py

Line 772 in 23a378f

bg_img = super().get_input(batch, "bg_concat")

[uishd666]

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 ?

[uishd666]

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.