ox-vgg / vgg_face2

645 stars 114 forks source link

All the download links are dead #52

Open ivanlen opened 3 years ago

ivanlen commented 3 years ago

All the links targeting https://www.robots.ox.ac.uk/~vgg/data/vgg_face2/ are dead.

Neither the models nor the dataset links are online.

zakerifahimeh commented 3 years ago

Hi Dear Can I have a pre-trained model?

dwhite54 commented 3 years ago

Looks like some model weights can be found here: https://www.robots.ox.ac.uk/~albanie/pytorch-models.html

I've pasted the relevant table here:

model (weights) training data (pretraining) IJB-A IJB-B IJB-C
resnet50_scratch_dag (weights) VGGFace2 0.895 ± 0.019 0.878 0.900
resnet50_ft_dag (weights) VGGFace2 0.908 ± 0.017 0.891 0.913
senet50_scratch_dag (weights) VGGFace2 (Ms-Celeb-1M) 0.904 ± 0.020 0.888 0.910
senet50_ft_dag (weights) VGGFace2 (Ms-Celeb-1M) 0.921 ± 0.014 0.908 0.927

(I've not verified these yet)

GaryGky commented 3 years ago

How about identity-meta.csv? I can not find it anywhere.

ucalyptus2 commented 2 years ago

cc: @WeidiXie @lishen-shirley

zengxijuan commented 2 years ago

Hi, I want to download the VGGFACE2 data, could you please help me? Maybe any new link that is accessible?

neeek2303 commented 2 years ago

Hi, @dwhite54. This links don't work, could you please update them?

dwhite54 commented 2 years ago

@neeek2303 sorry, it seems the links were removed by that website's owner. I don't have any other way to obtain those, so I guess you'll have to figure out how to convert the other listed weights here: https://github.com/ox-vgg/vgg_face2#trained-models

sbelharbi commented 1 year ago

hi, did anyone find working links for the pretrained weights? https://github.com/ox-vgg/vgg_face2/issues/57

thanks

adamhrv commented 1 year ago

It appears this dataset is deprecated but VGG doesn't seem to respond to any questions about its legal status.

sbelharbi commented 1 year ago

good find https://github.com/ox-vgg/vgg_face2/issues/54. i wasnt aware of the legal issues. that explains it. it seems to go back to 2020. thanks