search

THUDM / ImageReward

[NeurIPS 2023] ImageReward: Learning and Evaluating Human Preferences for Text-to-image Generation
Apache License 2.0
1.1k stars 60 forks source link

Finetuned model weights #90

Closed rohitrango closed 2 weeks ago

rohitrango commented 1 month ago

Hi,

Thank you for releasing code for the work! Have you released the weights for any finetuned model in the paper?

Thanks.

xujz18 commented 1 month ago

Thanks for your attention! You can construct your own prompt set to get a model that better meets your needs.

84

rohitrango commented 1 month ago

Why does the script scripts/train_refl_sdxl.sh require images too (train_refl.sh doesnt)?

Cominclip commented 4 weeks ago

Why does the script scripts/train_refl_sdxl.sh require images too (train_refl.sh doesnt)?

I meet the same question, do you have solved it? i don't understand what the images refer to..

xujz18 commented 3 weeks ago

This is code contributed by a community friend who supposedly added the pre-training loss function mentioned in the paper. You can refer to https://github.com/THUDM/ImageReward/pull/89