search

wangzhuo2019 / SSAN

Domain Generalization via Shuffled Style Assembly for Face Anti-Spoofing, CVPR2022.
MIT License
93 stars 22 forks source link

solved #12

Open TooManyMatts opened 2 years ago

TooManyMatts commented 2 years ago

solved.

fradino commented 1 year ago

I have the same question with you, but I don't know why.

Hi,

thank you very much for your work and for releasing the code :) I wanted to reproduce the results of your SSAN-R method on the OCIM benchmark (Table 2 in your paper) and got the following results:

image

As you can see, results on protocols "O&C&I to M", and "O&M&I to C" are quite similar, but results on protocols "O&C&M to I" and "I&C&M to O" have AUC below 90 which is not that good. Do you have any ideas where would that gap in performance be coming from?

Perhaps the train-test split is slightly different. If that is the case, could you please share the train_list_video.txt and train_list_video.txt files?

Thanks again for your work, Matt.

alvarobasi commented 1 year ago

Hi! I would be very grateful if you can explain how did you solved the issue. I'm having the same problem... Thanks!