Full refl_data.json for REFL

THUDM / ImageReward

[NeurIPS 2023] ImageReward: Learning and Evaluating Human Preferences for Text-to-image Generation

Apache License 2.0

1.12k stars 62 forks source link

Full refl_data.json for REFL #61

Open fuqianya opened 1 year ago

fuqianya commented 1 year ago

Hi,

Could you kindly to release the full refl_data.json for REFL? How many data do you used for REFL?

xujz18 commented 1 year ago

As we said in the "4.2 ReFL" section of the paper,

All fine-tuning methods use the same dataset as the pre-training dataset or generated dataset (both contain 20,000 samples).

The full refl_data.json contains pre-training dataset which need to download images (ReFL dataset just need prompts), so we didn't release pre-training dataset to simplify. Additionally, we suggest that you can try your own pre-training dataset which may lead to interesting results.