lisadunlap / alia Goto Github PK

Hey!

Hope all is well. After a while, came back to this repo.
A few questions:

1. What is the waterBirds dataset split? it seems to not be the full dataset. How can one get the split used in the paper?
2. Was the planes dataset completely removed? I suddenly don't see it in the paper. Is it possible to get the info and split of it (like the metadata.csv for waterbirds)?
3. I think that the planes dataset split is indeed interesting, and worth of publication. It seems like a much more realistic contextual bias that the WaterBirds dataset (which is literally not real)

Thanks!

1. Can I assume that the hyperparams noted in the paper are the ones I should use in order to recreate the results? because there is a difference between the corresponding config and what is noted in the paper. For example, Planes datasets LR in the config: 0.0005. In the paper it is 0.001. Same question for CUB.
2. Batch size is not noted anywhere, except in the config; is batch size = 128 was used for all datasets?

Thanks!

Hi,
Congratulations..!! This is amazing work.

I wanted to try this ALIA model on a dataset which I have. Can you please let me know steps of how to run this model on our own dataset..

Thank you

When I run the img2img.py code, it reports an error:

Firstly, Thank you for your work!
According to the README, recreating the CUB experiment is possible. However, it's not currently because the models folder is not yet pushed.
I assume that is a simple mistake and that it was just forgatten :)