search

brandontrabucco / da-fusion

Effective Data Augmentation With Diffusion Models
MIT License
221 stars 18 forks source link

Concept Erasure on PASACL dataset #11

Open tsWen0309 opened 1 year ago

tsWen0309 commented 1 year ago

Hi, @brandontrabucco

Sorry to bother you again. Which pre-trained concept erasure weights do you use in the implementation? Weights provided on https://erasing.baulab.info/weights/esd_models/ didn't involve any weights that were trained on pasacl dataset.

brandontrabucco commented 1 year ago

Hello again Flu0XeT1n,

We fine-tuned our own set of models for the PASCAL task, which I will upload to this google drive folder: https://drive.google.com/drive/folders/11aRHMWqD6PjslHKbdX6m2hKcua3n0Ia9?usp=sharing

If you check back in a couple days, the models should be uploaded. There is one checkpoint per class, where that class is erased.

tsWen0309 commented 1 year ago

Thank you so much

tsWen0309 commented 1 year ago

hey have u uploaded the erased model yet? The google drive is still empty.

AlvinHan123 commented 2 months ago

Thanks for your work, but the folder is empty

Hello again Flu0XeT1n,

We fine-tuned our own set of models for the PASCAL task, which I will upload to this google drive folder: https://drive.google.com/drive/folders/11aRHMWqD6PjslHKbdX6m2hKcua3n0Ia9?usp=sharing

If you check back in a couple days, the models should be uploaded. There is one checkpoint per class, where that class is erased.

Thanks for your work, but the folder is empty