You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
@justinpinkney hello, excuse me,
the first question is:1. for the command:"pipe:ssh -i ~/.ssh/id_rsa [email protected] cat /mnt/data_rome/laion/improved_aesthetics_6plus/ims",why the tar_base can't be visited and the dataset can't be visited.Do I need download the dataset to local?
the second question is : 2.I want to finetune image variation,I want to use my own data,but when I refer to your code,the sd-image-condition-attn-finetune.yaml is :
data:
target: main.DataModuleFromConfig
params:
batch_size: 4
num_workers: 2
num_val_workers: 0
train:
target: ldm.data.simple.FolderData
params:
root_dir: /data/xuzhenhao/ControlNet1/training/texture
ext: jpg
image_transforms:
- target: torchvision.transforms.Resize
params:
size: 256
interpolation: 3
- target: torchvision.transforms.RandomCrop
params:
size: 256
validation:
target: ldm.data.simple.FolderData
params:
root_dir: /data/xuzhenhao/ControlNet1/training/texture
ext: jpg
image_transforms:
- target: torchvision.transforms.Resize
params:
size: 256
interpolation: 3
- target: torchvision.transforms.CenterCrop
params:
size: 256
however,I find that in the FolderData class,why only have a data["image"],I think training need two image condition,one for clip encoder,one for noise and groundtruth, and why need data["txt"],Why need text for image variation?
I sincerely thank you,hope a detailed answer.
The text was updated successfully, but these errors were encountered:
@justinpinkney hello, excuse me,
the first question is:1. for the command:"pipe:ssh -i ~/.ssh/id_rsa [email protected] cat /mnt/data_rome/laion/improved_aesthetics_6plus/ims",why the tar_base can't be visited and the dataset can't be visited.Do I need download the dataset to local?
the second question is : 2.I want to finetune image variation,I want to use my own data,but when I refer to your code,the sd-image-condition-attn-finetune.yaml is :
data:
target: main.DataModuleFromConfig
params:
batch_size: 4
num_workers: 2
num_val_workers: 0
train:
target: ldm.data.simple.FolderData
params:
root_dir: /data/xuzhenhao/ControlNet1/training/texture
ext: jpg
image_transforms:
- target: torchvision.transforms.Resize
params:
size: 256
interpolation: 3
- target: torchvision.transforms.RandomCrop
params:
size: 256
validation:
target: ldm.data.simple.FolderData
params:
root_dir: /data/xuzhenhao/ControlNet1/training/texture
ext: jpg
image_transforms:
- target: torchvision.transforms.Resize
params:
size: 256
interpolation: 3
- target: torchvision.transforms.CenterCrop
params:
size: 256
however,I find that in the FolderData class,why only have a data["image"],I think training need two image condition,one for clip encoder,one for noise and groundtruth, and why need data["txt"],Why need text for image variation?
I sincerely thank you,hope a detailed answer.
The text was updated successfully, but these errors were encountered: