HanSolo9682 / CounterCurate

This is the implementation of CounterCurate, the data curation pipeline of both physical and semantic counterfactual image-caption pairs.
https://countercurate.github.io/
14 stars 0 forks source link

Combined Data Traning #6

Open lezhang7 opened 5 days ago

lezhang7 commented 5 days ago

Hi Authors,

Thank you for your great work! May I ask have you tried to combine all data and train a CLIP model on top of it? We want to make this dataset as a baseline and want to reproduce and achieve a fair comparison.

Best

HanSolo9682 commented 3 days ago

Hi, we actually have not combined all data together. You are welcome to test it out! One suggestion though is to see if you want the entire positions and counting dataset to be combined with the attributes dataset. The portions will affect the results, and make sure models don’t overfit to one of the tasks.