Prepare Dataset¶

Finetune¶

Prepare metadata.csv and images.

The folder structure is

data/example
├── color.jpg
├── ...
└── metadata.csv

Example of metadata.csv.

file_name,text
color.jpg,"a dog"

Fix dataset config.

train_dataloader = dict(
    ...
    dataset=dict(
        ...
        dataset="data/example",
        image_column="file_name",
        csv="metadata.csv",
        ...
        )
    ...
)

Run training.

DreamBooth¶

Prepare images.

The folder structure is

data/example
├── dog1.jpg
├── ...
└── dog5.jpg

Fix dataset config.

train_dataloader = dict(
    ...
    dataset=dict(
        ...
        dataset="data/example",
        )
    ...
)

Run training.

ControlNet¶

Prepare metadata.csv and images.

The folder structure is

data/example
├── images
|   └──  color.jpg
├── condition_images
|   └──  color_keypoint.jpg
└── metadata.csv

Example of metadata.csv.

file_name,conditioning_image,text
images/color.jpg,condition_images/color_keypoint.jpg,"a dog"

Fix dataset config.

train_dataloader = dict(
    ...
    dataset=dict(
        ...
        dataset="data/example",
        image_column="file_name",
        condition_column="conditioning_image",
        caption_column="text",
        csv="metadata.csv",
        ...
        )
    ...
)

Run training.