Clarification on Human Preference JSON File

yk7333 / d3po

[CVPR 2024] Code for the paper "Using Human Feedback to Fine-tune Diffusion Models without Any Reward Model"

https://arxiv.org/abs/2311.13231

MIT License

168 stars 14 forks source link

Clarification on Human Preference JSON File #17

Open jxxtin opened 1 month ago

jxxtin commented 1 month ago

I am planning to create a human preference JSON file, but I would like some clarification on the following:

After running accelerate launch scripts/sample.py, 140 images are generated in the data folder.

Should I score each image individually in the form of a dictionary, or is a single list of 140 scores sufficient?

if so, how should dimensions be configured?

It would be helpful if I could refer to an example of the JSON structure.

Thanks! @yk7333

yk7333 commented 3 weeks ago

You can use a list, for example, if the default is to generate 7 images for each prompt and a total of 2 prompts are sampled, then the JSON file can be in the format of [[0,0,1,0,0,1,1],[1,0,0,1,0,0,0]].