search

ceshine / yt8m-2019

7th place solution to The 3rd YouTube-8M Video Understanding Challenge
MIT License
37 stars 13 forks source link

whitening #5

Closed ibrahimrazu closed 4 years ago

ibrahimrazu commented 4 years ago

Hi, Thanks for sharing your awesome code. I am just curious, did you implement the reverse whitening mentioned in the paper?

Thanks!

ceshine commented 4 years ago

Not sure which paper you're refereing to. My paper does not mention whitening, so I'd guess that it did not get implemented.

ibrahimrazu commented 4 years ago

Thanks a lot for your reply. Yes, i also understood later on that whitening was not implemented. however, i have another question. for DBoF and NextVLAD training, you used steps and only one epoch. is that the standard?

ceshine commented 4 years ago

I definitely wouldn't call it the standard. It's just a hyper-parameter that I found working well enough given the limited computing resources. You're welcome to try other settings if your hardware configuration is different from mine (e.g., multiple GPUs).

EDIT: grammar

ibrahimrazu commented 4 years ago

i see, thanks a lot for your nice repo. it helps me a lot :)

ceshine commented 4 years ago

Glad to be of help!