search

timothybrooks / instruct-pix2pix

Other
6.41k stars 540 forks source link

How to prepare own dataset for training. #142

Open lxrswdd opened 2 weeks ago

lxrswdd commented 2 weeks ago

I have a set of source images and I have a set of target images, they are all stored locally. I wish to edit the source images to target images with their respective prompts.

However, the 'https://instruct-pix2pix.eecs.berkeley.edu/gpt-generated-prompts.jsonl' tutorial shows something I don't understand. Why do we have an output here as a text string?

{"input": "\"Dreama Tolle Perry \u00bb Artist and Writer \u00bb \"\"Tuscany Outside My Door\"\"\"", "edit": "Move the Tuscany to a different location", "output": "\"Dreama Tolle Perry \u00bb Artist and Writer \u00bb \"\"Tuscany in Alaska Outside My Door\"\"\"", "url": "https://i.pinimg.com/236x/d4/7d/4a/d47d4a43aeedae07fdac54f72e28cc71--flower-paintings-art-paintings.jpg"}

Should not we have a JSON that tells the model the directory of source images and target images with their prompts only?