finetuning pretrained model using the custom dataset

[mohammadshahabuddin]

Hi, Thank you for your great work. I am working on finetuning the pre-trained model using a custom dataset. Could you please let me know how I should organize the paired images and editing instructions during finetuning? I tried to download your provided datasets to get an idea but those datasets are more than 700GB. Could you provide a tiny version of the dataset which will contain only a few samples for reference? Thank you in advance.

[LIKP0]

Hi, you can access to clip-filtered-dataset and just download part of data (14GB), then you can check and find the pattern of the dataset.

Remember to download seeds.json.

[mohammadshahabuddin]

Thank you very much!

[unmo]

Hello. I have a question. What is seeds.json used for?

[LIKP0]

Hello @unmo. The seeds.json contains an array like [ [A, [B, C, D, E] , [A, [B, C, D, E], ...] where "A"s indicates the sub directory name in the dataset and "BCDE" are the image pair names in sub directory "A". Notice that the two images for "B" are represented as "B_0.jpg" (before edit) and "B_1.jpg" (after edit), which should be edited according to the prompt.json file in "A".

[unmo]

I understood the structure. Thank you very much.