Data to reproduce clip_coco.py

[j-min]

Hi, thanks for sharing your code.
I'd like to use the clipping method you proposed. However, I couldn't find the male_word_bank.txt, female_word_bank.txt, coco_precomp/testall_ids.txt, coco_testall_caps.txt, data/gender_coco.npy files used in the clip_coco.py.
I also checked the https://github.com/mjskay/gender-in-image-search repo mentioned in README, but I couldn't find the data in the repo. Could you please guide and share the data needed to run clip_coco.py? Thanks!

[Faldict]

Hi, the male_word_bank and female_word_bank are the list of masculine and feminine words, respectively. You may personalize them. testall_ids.txt and coco_testall_caps.txt are available from the original COCO dataset. The gender_coco.npy is the annotated gender attirbutes.

[j-min]

I see, where can I get the gender_coco.npy?
Actually, could you please provide the mis indicating which dimensions of CLIP features have been clipped?
I'd like to experiment with if clipping the feature dimensions helps to reduce gender bias on another dataset.

[TobiasLee]

I see, where can I get the gender_coco.npy? Actually, could you please provide the mis indicating which dimensions of CLIP features have been clipped? I'd like to experiment with if clipping the feature dimensions helps to reduce gender bias on another dataset.

Come to the same issue, and looking forward to the release of gender_coco.npy and the mis!

[itsyoavshalev]

Hi, can you please upload male_word_bank, female_word_bank and gender_coco.npy you used in your experiments?

Thanks.

[timqqt]

Hi, I believe the word banks are in the Section A of their Appendix. But are they all the ones used in the paper?

[Faldict]

Hi, I believe the word banks are in the Section A of their Appendix. But are they all the ones used in the paper?

Yes, these words are the ones used for generating the gender attributes. However, you may need a keen eye on the unexpected words that may disclose the gender information.